WO2011021588A1 - 移動体軌跡識別システム - Google Patents
移動体軌跡識別システム Download PDFInfo
- Publication number
- WO2011021588A1 WO2011021588A1 PCT/JP2010/063770 JP2010063770W WO2011021588A1 WO 2011021588 A1 WO2011021588 A1 WO 2011021588A1 JP 2010063770 W JP2010063770 W JP 2010063770W WO 2011021588 A1 WO2011021588 A1 WO 2011021588A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- trajectory
- identification information
- likelihood
- hypothesis
- information
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
Definitions
- the present invention relates to a moving body trajectory identification system that tracks the position of a moving body while discriminating identification information of the moving body and determines which moving body the recognized trajectory is.
- the present invention also relates to a moving object locus identification method and a moving object locus identification program.
- tracking of a moving object such as a person or an object is realized by associating the position information and identification information of the moving object. That is, the mobile object can be uniquely identified and tracked by associating the position information of the mobile object with the identification information.
- Various techniques have been developed for tracking a moving object (see, for example, Patent Document 1 and Patent Document 2).
- Patent Document 1 discloses a person tracking device, in which a person in a region is photographed with a sensor (visible camera), and feature information of the person is extracted based on a video signal output from the sensor. This person tracking device learns by associating person identification information and feature information, and identifies the person ID from the newly extracted feature information and the learning result. The person tracking device outputs the identified person ID and the trajectory information of the person as tracking information. If tracking of a person fails in the middle, the person tracking device identifies the person's ID again from the characteristic information and learning result of the person whose tracking has been resumed. That is, when the tracking of a person is interrupted, the person tracking device identifies the person ID again from the feature information extracted when the tracking is resumed.
- Patent Document 2 discloses a moving body tracking system including a monitoring camera for photographing a predetermined space. This moving body tracking system obtains position information by identifying a moving body based on the output of the monitoring camera, and manages the position information and first identification information unique to each moving body in association with each other. Further, the mobile body tracking system reads the second identification information unique to each mobile body, identifies each mobile body, and manages the positional information and the second identification information in association with each other. The mobile tracking system includes a position management table that manages position information and third identification information in association with each other. Furthermore, the moving body tracking system includes a position estimation table that manages the moving body recognized within a predetermined error range at the same time in association with the second identification information and the third identification information. The moving body tracking system tracks the moving body with reference to the position management table and the position estimation table. That is, the moving body tracking system tracks the moving body by integrating detection information obtained by a plurality of sensors.
- JP 2008-299742 A (see paragraphs 0006 and 0008 to 0013) Japanese Patent Laying-Open No. 2005-31955 (see paragraphs 0008 and 0009)
- a camera can track a moving body, but since it does not perform identification of the moving body, it cannot grasp which moving body is being tracked only by the camera.
- individual identification devices such as RFID (Radio Frequency IDentification) readers at various locations in the monitoring area, it is possible to grasp which mobile body has added the vicinity of the individual identification device, but it has moved away from the individual identification device. It is impossible to grasp the trajectory of the moving object at the place.
- RFID Radio Frequency IDentification
- the person tracking device disclosed in Patent Document 1 combines a camera and extraction of person feature information
- the moving object tracking system disclosed in Patent Document 2 combines a camera and a plurality of sensors to track a moving object. Is realized.
- tracking of a specific moving object is interrupted.
- tracking of the moving body can be realized when the feature information can be extracted or the identification information can be extracted by the sensor.
- the tracking is resumed based on the newly acquired feature information and identification information, and the tracking accuracy of the moving object is lowered.
- tracking of the moving object is resumed based on the newly acquired small amount of information, and therefore it is impossible to accurately determine which moving object is the recognized locus.
- the present invention has been made in view of the above circumstances, and an object thereof is to provide a mobile object trajectory identification system that can determine which mobile object a trajectory is based on identification information. .
- the present invention also provides a moving object locus identification method and a moving object locus identification program.
- a mobile object trajectory identification system includes a trajectory link candidate generation unit that generates a trajectory link candidate that is a combination of trajectories of a mobile object detected in a past fixed time, and identification of a mobile object detected in a past fixed time.
- a hypothesis generator that generates a set of trajectory link candidates / identification information pairs by combining information with trajectory link candidates, and generates a set of trajectory link candidates / identification information pairs that satisfy a predetermined condition as a hypothesis, and individual hypotheses
- the identification information likelihood indicating the likelihood that the identification information is detected for the trajectory represented by the trajectory link candidate is calculated, and the trajectory link candidate / identification information is calculated.
- the likelihood that the identification information likelihood for the selected hypothesis is calculated by integrating the identification information likelihood for each pair, and the maximum likelihood hypothesis is estimated based on the identification information likelihood of each hypothesis It comprises a calculation unit.
- a trajectory link candidate that is a combination of trajectories of a mobile object detected in a past fixed time is generated, Hypotheses that combine the identification information of mobile objects detected in the past certain time with the trajectory link candidate to generate a set of trajectory link candidates / identification information pairs and satisfy the predetermined condition with the set of trajectory link candidates / identification information pairs Generate an identification information likelihood indicating the likelihood that the identification information is detected for the trajectory represented by the trajectory link candidate in the trajectory link candidate / identification information pair belonging to the selected hypothesis.
- the identification information likelihood for the selected hypothesis is calculated by integrating the identification information likelihood for each trajectory link candidate / identification information pair. To estimate the most likely hypothesis on the basis of time.
- a probability map in which the detection probability of the identification information of the moving object according to the position coordinates in the tracking area may be referred to.
- the trajectory / identification information likelihood which is the likelihood that the trajectory of the moving object is associated with the identification information, may be referred to. Or you may refer to the attribute information and movement information of a moving body.
- the maximum likelihood hypothesis may be estimated based on a trajectory link likelihood that is a likelihood that trajectories included in the trajectory link candidate in the trajectory link candidate / identification information pair are linked.
- a hypothesis group may be generated by selecting trajectory link candidates corresponding to trajectory link likelihoods equal to or greater than a threshold.
- the location of the obstacle in the tracking area and the moving time when the moving object passes through the obstacle are used as environment information, or the environment represents the moving time of the moving object along the passable path in the tracking area. Information may be used.
- the actual travel time from the disappearance detection time of the trajectory in the trajectory link candidate to the next appearance detection time of the trajectory is calculated, and the travel time of the moving object is estimated based on the environment information.
- a trajectory link candidate whose estimated travel time is longer than the actual travel time is excluded from the trajectory link candidate group to be hypothesized.
- the maximum likelihood hypothesis is estimated by combining a plurality of parameters, so that the disappearing portion of the trajectory can be estimated with high accuracy even when the trajectory of the moving object is interrupted. That is, even when the tracking of the track of the moving body frequently occurs, the tracking of the moving body can be connected with high accuracy, and thus the tracking performance can be greatly improved.
- FIG. 1 It is a figure which shows a mode that the tracking area
- FIG. 1 It is a figure which shows the estimation procedure of the locus
- FIG. 2 It is a figure which shows the estimation procedure of the locus
- FIG. It is a figure which shows the estimation procedure of the locus
- FIG. It is a block diagram which shows the structure of the mobile body locus
- FIG. 10 is a block diagram which shows the structure of the likelihood calculation part in a mobile body locus
- 10 is a flowchart illustrating an operation of the mobile object locus identification apparatus according to the second embodiment.
- 10 is a flowchart illustrating an operation of the mobile object locus identification apparatus according to the second embodiment.
- It is a block diagram which shows the structure of the mobile body locus
- 10 is a flowchart illustrating an operation of the mobile object locus identification apparatus according to the third embodiment.
- FIG. 10 is a flowchart illustrating an operation of the mobile object locus identification apparatus according to the third embodiment. It is a figure for demonstrating the effect of likelihood integration in Example 3.
- FIG. It is a block diagram which shows the structure of the moving body locus
- 10 is a flowchart illustrating an operation of a moving body trajectory identification device according to a fourth embodiment. 10 is a flowchart illustrating an operation of a moving body trajectory identification device according to a fourth embodiment.
- trajectory identification device. 10 is a flowchart illustrating an operation of a moving body locus identification apparatus according to a fifth embodiment. 10 is a flowchart illustrating an operation of a moving body locus identification apparatus according to a fifth embodiment. It is a block diagram which shows the structure of the mobile body locus
- FIG. 12 is a flowchart illustrating an operation of a moving object locus identification apparatus according to the sixth embodiment.
- 12 is a flowchart illustrating an operation of a moving object locus identification apparatus according to the sixth embodiment.
- It is a block diagram which shows the structure of the moving body locus
- Example 7 it is the figure which showed typically the rejection determination process of the locus
- Example 7 it is a figure which shows typically the rejection determination process of the locus
- 12 is a flowchart illustrating an operation of a moving body trajectory identification device according to a seventh embodiment. It is a block diagram which shows the minimum structure of the mobile body locus
- FIG. 1 is a block diagram showing a moving body trajectory identification system according to Embodiment 1 of the present invention.
- This moving body trajectory identification system includes a position information detecting device 1, an identification information detecting device 2, a moving body trajectory identifying device 3, and an identification result output device 4.
- the moving object trajectory identification system identifies which moving object is the trajectory of the moving object in the predetermined tracking area 50. That is, it is possible to recognize where the moving object P exists.
- the moving body P moves freely in the tracking area 50. Further, the moving body P may move outside the tracking area 50.
- the kind of moving body is not specifically limited, A human, an animal, or a thing may be sufficient.
- the position information detection device 1 detects the position of the moving body in the tracking area 50. In the following description, it is assumed that the position in the tracking area 50 is detected by two-dimensional coordinates.
- the position information detection apparatus 1 outputs a set of two-dimensional position coordinates (hereinafter simply referred to as “position coordinates”) of a moving object, its detection time, and a locus ID in the tracking area 50 to the moving object locus identification information 3. To do.
- the detected position coordinates may be referred to as position information.
- the locus ID constitutes identification information, and is assigned to determine that the position information of the same moving body is the same as the position information of the moving body that has been successfully detected continuously. That is, the position coordinate group to which the same locus ID is assigned represents the locus of the same moving object.
- the trajectory ID is expressed in a number format (that is, trajectory number), but the trajectory ID may be expressed in a format other than the number.
- the position information detection device 1 may be any device that can detect the position coordinates of the moving body in the tracking area 50, specify the detection time, and assign the trajectory number.
- the position information detection apparatus 1 can be realized by a camera, a floor pressure sensor, a GPS (Global Positioning System), or the like.
- the position information detection device 1 is desirably installed so that the entire tracking area 50 can be detected without a blind spot, but a blind spot that may be partially undetectable may occur. This is because the moving object trajectory identification device 3 can connect the trajectories fragmented and identify which moving object the trajectory is.
- the identification information detection apparatus 2 acquires the identification information of the moving object in the tracking area 50. However, even if there is a moving object in the tracking area 50, the identification information detecting device 2 does not always detect the identification information, and the probability that the identification information of the moving object can actually be detected is the movement in the tracking area 50. It depends on the two-dimensional position coordinates of the body. For example, the detection probability of the identification information of the mobile object existing at a position close to the identification information detection device 2 is high, while the detection probability of the identification information of the mobile object existing at a position away from the identification information detection device 2 is low.
- the identification information detection device 2 uses one identification information detection apparatus 2, but a plurality of identification information detection apparatuses 2 may be installed so as to cover the entire tracking area 50.
- the identification information detection device 2 is assigned in advance with an identification information detection device ID for uniquely identifying it.
- the identification information detection device ID is used to determine which identification information detection device 2 has detected the identification information.
- the identification information detection device ID is represented by a number format (that is, identification information detection device number), but may be represented by a format other than a number.
- the identification information detection device 2 outputs the detected identification information, its detection time, and its identification information detection device number as a set to the mobile object locus identification device 3.
- the identification information is set to “none” and the time and the identification information detection device number are output to the moving body locus identification device 3.
- the identification information detection device 2 may not output identification information to the moving body locus identification device 3, so that the moving body locus identification device 3 determines that the identification information is not detected at that time.
- the identification information detection device 2 may be any device that can detect identification information and specify the detection time and the identification information detection device number.
- an RFID reader can be employed as the identification information detection device 2.
- the moving body is a person, a person's face, fingerprint, vein, or the like can be used as identification information, and a reading device for such identification information can be employed as the identification information detection apparatus 2.
- the detection areas of the identification information detection devices 2 may be installed so as to overlap each other. Or you may install so that a detection area may not mutually overlap.
- the position information detection device 1 detects the position coordinates of the moving body
- the identification device detection device 2 detects the identification information.
- the moving object trajectory identification device 3 buffers the position coordinates and the identification information for a certain period of time. You may make it use the position coordinate and identification information which were accumulate
- the moving body trajectory identification device 3 uses the position information sent from the position information detection device 1 and the identification information sent from the identification information detection device 2 to determine which moving body the moving body trajectory is. judge.
- the moving body trajectory identification information 3 sets a combination of moving body trajectories corresponding to the trajectory number as a trajectory link candidate, and determines a combination of identification information and trajectory link candidates that are most likely. For this reason, it can be said that the trajectory link candidate is a link of fragmented moving object trajectories.
- FIG. 2 is a block diagram showing the configuration of the moving object trajectory identification device 3.
- the moving body trajectory identification device 3 is connected to the position information detection device 1, the identification information detection device 2, and the identification result output device 4.
- the mobile object trajectory identification device 3 includes a trajectory link candidate generation unit 31 that performs a process of connecting fragmented mobile object trajectories, a hypothesis generation unit 32 that generates a set of trajectory link candidates and identification information as a hypothesis, and the likelihood of the hypothesis. And a likelihood calculation unit 33 that specifies a hypothesis with the highest likelihood and a map storage unit 37 are provided.
- a set of sets of trajectory link candidates and identification information is determined. This means that the correlation between the detected moving object trajectory and the moving object P in the tracking area 50 has been determined.
- the trajectory link candidate generation unit 31 receives a set of position information (that is, position coordinates), its detection time, and a trajectory number from the position information detection apparatus 1.
- a set of position coordinates which are position coordinates to which the same locus number is assigned and which are detected in a time zone between the current time and a time that is back by a certain time in the past represents one moving object locus.
- the trajectory link candidate generation unit 31 generates a combination of individual mobile trajectories, and outputs each combination to the hypothesis generation unit 32 as a trajectory link candidate.
- the trajectory link candidate generation unit 31 When the number of trajectories is N, the trajectory link candidate generation unit 31 generates everything from a trajectory link candidate including one trajectory to a trajectory link candidate including N trajectories. For example, when three trajectories are obtained, the trajectory link candidate generation unit 31 generates “trajectory 1”, “trajectory 2”, and “trajectory 3” as trajectory link candidates including one trajectory. Also, “trajectory 1, trajectory 2”, “trajectory 1, trajectory 3”, and “trajectory 2, trajectory 3” are generated as trajectory link candidates including two trajectories. Furthermore, “trajectory 1, trajectory 2, trajectory 3” is generated as a trajectory link candidate including three trajectories.
- the trajectory link candidate generation unit 31 generates seven trajectory link candidates for the three trajectories.
- the trajectory link candidate generation unit 31 does not generate trajectory link candidates by combining the same trajectories. For this reason, the trajectory number is not duplicated in the trajectory link candidate.
- a plurality of loci whose position coordinates are detected at the same time may be included in the loci connection candidates.
- the locus connection candidate generation unit 31 detects the position coordinates detected at the same time. Calculate the centroid point. For example, when the coordinate A of the trajectory 1 and the coordinate B of the trajectory 2 are detected at the same time, the trajectory link candidate generation unit 31 calculates the center of gravity of the coordinates A and B (that is, the midpoint between the two coordinates). Similar calculations are performed for other sets of coordinates detected at the same time.
- the coordinates are also referred to when calculating the identification information likelihood described later, but when there are a plurality of coordinates detected at the same time, the trajectory link candidate generation unit 31 calculates and refers to the center of gravity of the plurality of coordinates. To do.
- the hypothesis generation unit 32 inputs the identification information, its detection time, and the identification information detection device number from the identification information detection device 2, and holds the corresponding relationship. Further, the hypothesis generation unit 32 generates all combinations of trajectory link candidates and identification information, and outputs them to the likelihood calculation unit 33 as hypotheses.
- the hypothesis generation unit 32 detects the identification information detected by the identification information detection device 2 in a time zone between the trajectory link candidate input from the trajectory link candidate generation unit 31 and a time that is a predetermined time before the current time. Based on the above, all combinations of the trajectory link candidate and the identification information are generated.
- trajectory overdetection means that the trajectory of an object other than the tracking target is erroneously detected.
- overdetection of identification information means that an RFID reader installed in the tracking area is affected by radio wave interference even though a mobile object having an RFID tag exists outside the tracking area. It means a case where a person who is a moving body detects the identification information erroneously when passing in front of the identification information detection apparatus 2 using biometric authentication (face authentication).
- the missed detection of the trajectory means a case where the trajectory of the moving body that is moving in the location that becomes the blind spot of the position information detecting device 1 cannot be detected.
- Missing detection of identification information means that even though a mobile object having an RFID tag exists within the detection range of the RFID reader, the received radio wave is weakened due to the influence of radio wave interference, and the identification information is not detected.
- biometric authentication face authentication
- the hypothesis generation unit 32 also includes a set of information indicating a state in which there is no corresponding partner (that is, a corresponding trajectory link candidate or identification information) for all trajectory link candidates and identification information. Generate. Information indicating that there is no corresponding partner is represented by a character string “unknown”. For example, when the trajectory link candidate generation unit 31 generates trajectory link candidates T1, T2,..., Tn and the identification information detection device 2 detects the identification information ID1, ID2,. The generation unit 32 includes (T1, unknown), (T2, unknown), ..., (Tn, unknown), (unknown, ID1), (unknown, ID2), ..., (unknown, IDm). Is also generated and added to the set of the trajectory link candidate and the identification information. A set of a trajectory link candidate and identification information is referred to as a “trajectory link candidate / identification information pair”. A group including “unknown” also corresponds to a trajectory link candidate / identification information pair.
- the hypothesis generation unit 32 selects a set of trajectory link candidate / identification information pairs that satisfy a predetermined condition and sets it as a hypothesis.
- a plurality of sets of trajectory link candidate / identification information pairs that satisfy a predetermined condition may exist. For this reason, the hypothesis generation unit 32 extracts all sets of trajectory link candidate / identification information pairs that satisfy a predetermined condition, and sets each set as a respective hypothesis.
- the following three conditions are given as predetermined conditions for setting a set of trajectory link candidate / identification information pairs as a hypothesis.
- the first condition is that no trajectory duplication or identification information duplication occurs between trajectory link candidate / identification information pairs belonging to a hypothesis. For example, when identification information ID1 and ID2 are detected for four trajectories k1, k2, k3, and k4, a set of [ ⁇ (k1, k2), ID1 ⁇ , ⁇ (k1, k3, k4), ID2 ⁇ ] Then, since the trajectory k1 overlaps, the first condition is not satisfied. In the set [ ⁇ (k1, k2), ID1 ⁇ , ⁇ (k3, k4), ID1 ⁇ ], since the identification information ID1 is duplicated, the first condition is not satisfied.
- (k1, k2) represents a trajectory link candidate
- ⁇ (k1, k2), ID1 ⁇ represents a trajectory link candidate / identification information pair.
- the trajectory of one moving body is collected into one trajectory link candidate.
- the hypothesis is a candidate used to determine which moving body is a trajectory of each trajectory, and a hypothesis including duplication of trajectories and identification information is not an appropriate candidate, so the first condition is provided. .
- the second condition is that all individual trajectories identified by trajectory numbers are included in trajectory link candidates belonging to the trajectory link candidate / identification information pair.
- the set [ ⁇ (k1, k2), ID1 ⁇ ] includes only the trajectories k1 and k2 in the trajectory link candidate / identification information pair. And k3 and k4 are not included. Therefore, the second condition is not satisfied.
- the set [ ⁇ (k1, k2), ID1 ⁇ , ⁇ (k3, k4), ID2 ⁇ ] all the trajectories k1 to k4 are included in the trajectory link candidate / identification information pair. Meet the conditions.
- the third condition is that at least one trajectory link candidate / identification information pair whose identification information is not “unknown” is included in the set of trajectory link candidate / identification information pairs.
- the two items included in the set [ ⁇ (k1, k2), unknown ⁇ , ⁇ (k3, k4), unknown ⁇ ] Since the identification information is “unknown” in the trajectory link candidate / identification information pair, the third condition is not satisfied.
- the set [ ⁇ (k1, k2), ID1 ⁇ , ⁇ (k3, k4), unknown ⁇ ] includes a trajectory link candidate / identification information pair whose identification information is not “unknown”. Condition 3 is satisfied.
- the hypothesis generation unit 32 selects all sets of trajectory link candidate / identification information pairs that satisfy the first condition to the third condition as hypotheses. For example, when ID1 and ID2 are detected as identification information, one trajectory link candidate / identification information pair includes ID1, and another trajectory link candidate / identification information pair includes “unknown” as identification information. It is established as. Similarly, a set in which one trajectory link candidate / identification information pair includes ID2 and another trajectory link candidate / identification information pair includes “unknown” as identification information is also established as a hypothesis. Furthermore, one trajectory link candidate / identification information pair includes ID1, another trajectory link candidate / identification information pair includes ID2, and another trajectory link candidate / identification information pair includes “unknown” as identification information. Is also a hypothesis. A set including only trajectory link candidate / identification information pairs whose identification information is not “unknown” is also established as a hypothesis. The hypothesis generation unit 32 can select and cover all sets satisfying the first condition to the third condition as hypotheses.
- the map storage unit 37 stores an identification information likelihood map.
- the identification information likelihood map will be described later.
- the likelihood calculating unit 33 calculates two types of likelihood for each hypothesis.
- the first likelihood is the likelihood of connecting the trajectories included in the trajectory link candidates in the trajectory link candidate / identification information pair belonging to the hypothesis.
- the first likelihood is referred to as a trajectory link likelihood.
- the second likelihood is the likelihood that the identification information is detected for the activation represented by the trajectory link candidate in the trajectory link candidate / identification information pair.
- the second likelihood is referred to as identification information likelihood or identification information detection likelihood. If the trajectory represented by the trajectory link candidate is S and the identification information is O, the identification information likelihood is expressed as a conditional probability P (OiS).
- the identification information likelihood is calculated for each trajectory link candidate / identification information pair, and the result of integrating the identification information likelihood of the trajectory link candidate / identification information pair belonging to the hypothesis is the identification information likelihood of the hypothesis.
- FIG. 3 is a block diagram illustrating a configuration of a likelihood calculation unit according to the first embodiment.
- the likelihood calculation unit 33 includes a likelihood calculation control unit 330, a trajectory link likelihood calculation unit 331, and an identification information likelihood calculation unit 332.
- the likelihood calculation control unit 330 outputs a hypothesis for which the likelihood is not calculated to the trajectory link likelihood calculation unit 331 and the identification information likelihood calculation unit 332.
- the trajectory link likelihood calculation unit 331 calculates the trajectory link likelihood of the trajectory link candidate for each trajectory link candidate / identification information pair belonging to the hypothesis.
- the identification information likelihood calculation unit 332 calculates the identification information likelihood in the trajectory link candidate / identification information pair for each hypothesis.
- the identification information likelihood calculation unit 332 sets the hypothesis likelihood as a result of integrating the calculation results of the two kinds of likelihoods (that is, the trajectory link likelihood and the identification information likelihood), and selects a hypothesis having a high likelihood.
- the maximum likelihood estimation result shown is sent to the identification result output device 4.
- trajectory link candidates by the trajectory link likelihood calculation unit 331 .
- calculation when the trajectory link candidate is a combination of two trajectories will be described.
- the combination of trajectory link candidates there are a case where trajectories detected redundantly at the same time are combined and a case where trajectories detected independently at different times are combined.
- the trajectory link likelihood calculation unit 331 calculates the Euclidean distance between coordinates detected at the same time in each trajectory, and the trajectory link likelihood is higher as the Euclidean distance is shorter, and the trajectory link likelihood is lower as the Euclidean distance is longer. What is necessary is just to determine a locus
- the trajectory link likelihood is determined such that the trajectory link likelihood is higher as the average value of the Euclidean distance calculated for each set of coordinates is smaller and the trajectory link likelihood is lower as the average value is larger.
- the trajectory link likelihood is determined using the average value of the Euclidean distance between coordinates, but a variance value may be used instead of the average value. If the start time and end time of the simultaneous detection of coordinates between the two trajectories do not completely match, the above processing is performed for the time zone in which the coordinates of the two trajectories are detected simultaneously.
- the trajectory link likelihood calculating unit 331 may calculate a movement vector of each trajectory and calculate the cosine similarity of the movement vector as the trajectory link likelihood.
- the movement vector of each trajectory is a vector whose starting point is a coordinate when simultaneous detection of coordinates in two trajectories is started and whose end point is a coordinate when simultaneous detection of coordinates is ended.
- the trajectory link likelihood calculating unit 331 calculates the cosine similarity cos ⁇ by calculating the movement vectors V1 and V2 according to Equation 1, and determines the trajectory link likelihood.
- ⁇ V1, V2> means an inner product of the movement vectors V1 and V2
- ⁇ V1 ⁇ means a norm of the movement vector V1
- ⁇ V2 ⁇ means a norm of the movement vector V2.
- trajectory connection likelihood may be predetermined in the likelihood calculation part 33, and you may make it exclude from a process target hereafter about the locus
- trajectory link likelihood calculation unit 331 calculates the moving speed of the moving object based on the time and position coordinates when the trajectory disappears earlier and the time and position coordinates when the trajectory appears later.
- the trajectory link likelihood calculation unit 331 increases the trajectory link likelihood as the error between the movement speed set in advance as the moving model of the moving object and the actually calculated movement speed decreases, and increases as the error increases.
- the trajectory link likelihood is calculated so that the degree becomes low.
- a different likelihood function may be defined so that the trajectory link likelihood is likely to decrease.
- data on the moving speed at the time when the interruption of tracking occurs is collected in advance, and a set in which the error between the collected moving speed data and the moving speed of the moving model is a positive value, and a set that is a negative value
- the average value and variance value of errors are calculated for each set.
- a replica is generated for the positive error set, and the average is set to zero by adding the set with the sign converted to negative to the original set.
- a replica is generated for the set of negative errors, and the set whose sign is converted to positive is added to the original set to make the average zero.
- the distribution of errors included in each set is a normal distribution
- the average of the set of positive errors is ⁇ s
- the variance is ⁇ s
- the average of the set of negative errors is ⁇ m
- the variance is ⁇ m.
- the probability density function is defined by Equation 2.
- the trajectory link likelihood calculating unit 331 calculates a trajectory link likelihood by substituting a calculation error (that is, a value obtained by subtracting a moving speed set in advance for the moving model from the calculated moving speed) into Equation 2 as a variable x.
- a calculation error that is, a value obtained by subtracting a moving speed set in advance for the moving model from the calculated moving speed
- a trajectory link likelihood threshold value is determined in advance, and the trajectory link likelihood less than the threshold value is excluded from the processing target thereafter.
- the above processing is related to the trajectory link candidate being a combination of two trajectories.
- the trajectory link likelihood calculating unit 331 performs the following procedure.
- the trajectory link likelihood is calculated according to
- the trajectory link likelihood calculating unit 331 orders the trajectories included in the trajectory link candidates from the earliest disappearance time.
- the trajectory link likelihood calculation unit 331 calculates the trajectory link likelihood for each of two trajectories that are adjacent in order. That is, the trajectory link likelihood calculation unit 331 calculates the trajectory link likelihood for the first and second trajectories in the order of the trajectory disappearance time. Similarly, the trajectory link likelihood is calculated for the second and third trajectories with the trajectory disappearance time order, and the trajectory link likelihood is calculated for the third and fourth trajectories with the trajectory disappearance time order. Note that the above-described method is used for the calculation of the trajectory link likelihood related to two trajectories adjacent in order.
- the trajectory link likelihood calculating unit 331 adds up the calculated trajectory link likelihoods and calculates the power root of the number of times of integration. .
- This calculation result is set as a trajectory link likelihood related to a combination of three or more trajectories.
- the calculation of the power root is a normalization process for the trajectory link likelihood.
- the trajectory link likelihood calculation unit 331 sets the trajectory link likelihood of a trajectory link candidate that includes only one trajectory as a constant (for example, “1”).
- the identification information likelihood calculation unit 332 calculates the identification information likelihood for each hypothesis generated by the hypothesis generation unit 32.
- the identification information likelihood calculating unit 332 calculates the identification information likelihood with reference to an identification information probability map previously determined for each identification information detection device 2.
- the identification information probability map is a map in which the tracking area 50 is divided into grids, and the probability that the identification information is detected by the identification information detection device 2 in each divided cell is defined by a numerical value between 0 and 1.
- the identification information likelihood calculating unit 332 calculates the identification information likelihood in the trajectory link candidate / identification information pair. Since the identification information likelihood calculation unit 332 calculates the identification information likelihood by referring to the identification information probability map detected in the past certain period from the current time in the identification information probability map, it is only detected at the current time.
- the correspondence relationship between the trajectory and the identification information can be determined without depending on. For example, even if the detection probability of identification information at the present time is low, the correspondence between the trajectory and the identification information can be determined with high accuracy if the mobile object has passed through a place with a high detection probability in the past.
- the identification information likelihood calculation unit 332 calculates the identification information likelihood for one hypothesis according to the following procedure. First, the identification information likelihood calculation unit 332 extracts the trajectory link candidate / identification information pair included in the hypothesis. Then, the identification information likelihood calculation unit 332 calculates and accumulates the identification information likelihood for each trajectory link candidate / identification information pair. The number of trajectory link candidates / identification information pairs included in each hypothesis is not constant, and there are trajectory link candidates / identification information pairs for which the identification information likelihood is not calculated. Therefore, the identification information likelihood calculation unit 332 normalizes the integration result of the identification information likelihood calculated for each trajectory link candidate / identification information pair.
- the identification information likelihood calculation unit 332 When calculating the identification information likelihood for each trajectory link candidate / identification information pair, the identification information likelihood calculation unit 332 detects the detection time when the identification information of the trajectory link candidate / identification information pair of interest is detected, and the identification The identification information detection device number of the identification information detection device 2 that has detected the information is referred to. Further, the identification information likelihood calculation unit 332 selects an identification information probability map corresponding to the identification information detection device number. The identification information probability map for each identification information detection device number is stored in advance in the map storage unit 37. Note that the identification information likelihood calculation unit 332 acquires from the hypothesis generation unit 32 a correspondence relationship between the identification information, its detection time, and the identification information detection device number of the identification information detection device 2 that has detected the identification information.
- the identification information likelihood calculation unit 332 is provided from the trajectory link candidate generation unit 31 with the correspondence between the position coordinates of the moving object, its detection time, and the trajectory number. When the position coordinates of different trajectories included in the trajectory link candidate are detected at the same time, the coordinates of the barycentric point are also provided from the trajectory link candidate generation unit 31.
- the identification information likelihood calculation unit 332 refers to the position coordinates of the trajectory link candidate at the detection time of the identification information, determines which cell of the identification information probability map the position coordinates correspond to, and thus corresponds. The likelihood of identification information defined in the cell is acquired.
- the identification information likelihood calculating unit 332 specifies the cell corresponding to the barycentric point coordinate. Then, the identification information likelihood defined in the cell is acquired.
- the identification information likelihood calculation unit 332 integrates the identification information likelihoods obtained at all identification information detection times. This integration result is the identification information likelihood of the trajectory link candidate / identification information pair of interest.
- the identification information likelihood calculation unit 332 does not calculate the identification information likelihood for the trajectory link candidate / identification information pair for which the identification information likelihood cannot be calculated. That is, the identification information likelihood is not calculated for the trajectory link candidate / identification information pair whose identification information or trajectory link candidate is “unknown”.
- the identification information likelihood calculation unit 332 calculates the identification information likelihood for each trajectory link candidate / identification information pair belonging to the hypothesis, integrates it, and normalizes it to obtain the identification information likelihood of the entire hypothesis. Since the identification information likelihood for each trajectory link candidate / identification information pair corresponds to the product of the probability values defined in the identification information probability map, the integration result of the identification information likelihood for each trajectory link candidate / identification information pair is also It corresponds to the product of probability values. Therefore, the number of products of probability values specified in the identification information probability map is specified, and the probability value is calculated for the identification information likelihood integration result for each trajectory link candidate / identification information pair. Normalization is performed by taking the power root of the number of. By this normalization, even if the number of trajectory link candidate / identification information pairs belonging to a hypothesis is different, the identification information likelihood can be appropriately compared between the hypotheses.
- the identification information likelihood calculation unit 332 calculates the identification information likelihood for all hypotheses by repeating the above procedure. Further, the identification information likelihood calculation unit 332 integrates the trajectory link likelihood and the identification information likelihood for each hypothesis, thereby calculating the likelihood of the entire hypothesis. It can be said that the hypothesis having the highest likelihood optimally represents the relationship between the trajectory and the identification information. The identification information likelihood calculation unit 332 selects the hypothesis having the highest likelihood and outputs it to the identification result output device 4.
- the identification result output device 4 presents the correspondence between the trajectory and the moving object based on the hypothesis (that is, the maximum likelihood estimation result) sent from the identification information likelihood calculation unit 332.
- Each trajectory link candidate / identification information pair belonging to the hypothesis represents a correspondence relationship between the trajectory and the identification information, so which trajectory corresponds to which mobile body based on the trajectory link candidate / identification information pair in the hypothesis. Judgment is made and presented. At this time, the position coordinates of the current time or the trajectory of a past fixed time are also presented.
- the identification result output device 4 is configured by a display device, for example. That is, the identification result output device 4 includes a display, and displays the correspondence between the locus and the moving object on the display.
- the output mode of the identification result output device 4 is not limited to the display mode, and may be other output modes such as print output.
- the trajectory link candidate generation unit 31, the hypothesis generation unit 32, and the likelihood calculation unit 33 are, for example, moved It is realized by a CPU of a computer that operates according to a body locus identification program. That is, a program storage device (not shown) of a computer stores a moving body trajectory identification program, and a CPU reads the moving body trajectory identification program, and a trajectory link candidate generation unit 31, a hypothesis generation unit 32, and likelihood calculation according to the program. Each function of the unit 33 is realized.
- the trajectory link candidate generation unit 31, the hypothesis generation unit 32, and the likelihood calculation unit 33 may be realized by separate hardware.
- the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, and the identification information likelihood calculation unit 332 may also be realized by separate hardware.
- FIG. 6 shows the positions and trajectory numbers at times t1 to t9 for the moving object A and the moving object B that move in the tracking area 50.
- the trajectory of the moving object A corresponds to the trajectory 1 and the trajectory 3
- the trajectory of the moving object B corresponds to the trajectory 2 and the trajectory 4.
- the identification information of only the moving object A can be detected, the identification information of the moving object B is not registered, and the corresponding identification information does not exist.
- FIG. 7 shows the positions of the identification information detection device 2a and the identification information detection device 2b installed in the tracking area, and grid division for defining the identification information probability map, where the lower left corner is the origin (0, 0).
- An arbitrary position coordinate is represented by p (x, y).
- the coordinates of an arbitrary cell on the identification information probability map are represented by c (a, b).
- the number of cell divisions may be set arbitrarily, in FIG. 7, it is set in the range of 0 to 8 in the a direction and 0 to 4 in the b direction. That is, the identification information detection devices 2a and 2b are arranged at cell coordinates c (0, 4) and c (8, 0), respectively.
- each cell of the identification information probability map shown in FIG. 7 a probability value in the range of 0 to 1 is defined according to the detection characteristics of each identification information detection device.
- the detection probability value of the identification information in each cell is set for each of the two identification information detection devices 2a and 2b.
- FIG. 8 shows an identification information probability map of the identification information detection device 2a
- FIG. 9 shows an identification information probability map of the identification information detection device 2b.
- the hypothesis generation unit 32 inputs the identification information, the detection time, and the identification information detection device number from the identification information detection devices 2a and 2b, and holds the corresponding relationship.
- FIG. 10 shows a table for storing this correspondence. This table stores which identification information detection device number of the identification information detection device detects the identification information for each time. Here, the presence / absence of detection of identification information for each identification information detection apparatus at times t1 to t9 is indicated. For example, the identification information detection device 2a detects the identification information ID1 at times t1 and t2, while the identification information detection device 2b detects the identification information ID1 at times t8 and t9. A blank in the table of FIG. 10 indicates that the identification information is not detected by the identification information detection devices 2a and 2b. 11, FIG. 12, FIG. 13 and FIG. 14 are diagrams for explaining the procedure for acquiring the identification information detection probability value on the identification information probability map of the tracking area.
- the trajectory link candidate generation unit 31 acquires a set of trajectory number, detection time, and position coordinates from the position information detection device 1 (step S1).
- the position information detection apparatus 1 uses ⁇ 1, t1, p (30, 80) as a set of the trajectory number, the detection time, and the position coordinates, for example, for time t1.
- ⁇ 2, t1, p (25, 10) ⁇ are output to the trajectory link candidate generation unit 31 of the mobile trajectory identification device 3.
- a set of a trajectory number, a detection time, and a position coordinate is provided to the trajectory link candidate generation unit 31 at times t2, t3,.
- the trajectory link candidate generation unit 31 holds the position information output from the position information detection apparatus 1 for a certain past time from the current time (that is, the latest time when information is input).
- the trajectory link candidate generation unit 31 additionally stores the latest position information and deletes the oldest position information.
- the history is updated (step S2). For example, in a state where the trajectory link candidate generation unit 31 holds a set of trajectory number, detection time, and position coordinate for times t1 to tn ⁇ 1, the trajectory number, detection time, and position coordinate for time tn.
- the position information at the time tn is newly stored, while the set of the trajectory number, the detection time, and the position coordinate regarding the time t1 is discarded.
- a set of position coordinates that are paired with the same trajectory number represents one trajectory.
- the trajectory link candidate generation unit 31 When the number of trajectories is N, the trajectory link candidate generation unit 31 generates everything from a trajectory link candidate including one trajectory to a trajectory link candidate including N trajectories. That is, the trajectory link candidate generation unit 31 not only includes trajectory link candidates including two trajectories as described above, but also includes trajectory link candidates including one trajectory such as (trajectory 1), (trajectory 2),. , (Trajectory 1, trajectory 2, trajectory 3), (trajectory 1, trajectory 2, trajectory 4),..., (Trajectory 1, trajectory 2, trajectory 3, trajectory) As shown in 4), a trajectory link candidate including four trajectories is generated. A trajectory link candidate combining trajectories detected at the same time is also generated.
- the trajectory link candidate generation unit 31 also calculates the position coordinates of the barycentric point when one coordinate of a different trajectory is detected at the same time in the generated trajectory link candidate.
- the trajectory link candidate generation unit 31 sets the trajectory number, detection time, and position coordinate set (that is, the trajectory number, detection time, and position coordinate set obtained in the update process in step S2) by itself. Not only storing, but also outputting to the likelihood calculation unit 33.
- the barycentric point of the position information detected at the same time which is the position coordinate of a different locus is calculated, the barycentric point is also output to the likelihood calculating unit 33.
- the trajectory link candidate generation unit 31 outputs these pieces of information to the likelihood calculation unit 33 via the hypothesis generation unit 32.
- the hypothesis generation unit 32 acquires a set of detection time, identification information, and identification information detection device number from the identification information detection device 2 (step S4). That is, when the identification information detection apparatus 2 detects the identification information of the moving object, the identification information detection apparatus 2 provides the hypothesis generation unit 32 with the related information.
- the hypothesis generation unit 32 holds the correspondence relationship between the detection time, the identification information, and the identification information detection device number in the table shown in FIG.
- the hypothesis generation unit 32 holds the identification information acquired from the identification information detection device 2 in a certain past time from the current time (that is, the time when the latest information is input).
- the hypothesis generation unit 32 acquires the identification information and the identification information detection device number at the latest detection time
- the hypothesis generation unit 32 additionally registers them in the table shown in FIG. 10 and deletes the oldest identification information (step S5).
- the hypothesis generation unit 32 holds a set of detection times, identification information, and identification information detection device numbers at times t1 to tn ⁇ 1, detection time, identification information, and identification information detection at the latest time tn
- the identification information acquired for the time tn is registered in the table shown in FIG. 10, while the identification information for the time t1 is discarded.
- the hypothesis generation unit 32 not only stores the updated table content itself, but also outputs it to the likelihood calculation unit 33. Further, the hypothesis generation unit 32 uses the identification information stored in the table updated in step S5 and the trajectory link candidate input from the trajectory link candidate generation unit 31 in step S3, that is, the pair of trajectory link candidates, An identification information pair is generated (step S6). In the example shown in FIGS. 6 and 10, the hypothesis generation unit 32 uses ⁇ (trajectory 1, trajectory 3), ID1 ⁇ , ⁇ (trajectory 2, trajectory 4), ID1 ⁇ ,... Is generated.
- the trajectory link candidate includes two trajectories, but the number of trajectories included in the trajectory link candidate is not limited to two.
- step S6 the hypothesis generation unit 32 also generates a trajectory link candidate / identification information pair including “unknown” indicating that the corresponding partner does not exist. Accordingly, trajectory link candidate / identification information pairs such as ⁇ (trajectory 1, trajectory 3), unknown ⁇ , ⁇ (trajectory 2, trajectory 4), unknown ⁇ , ⁇ (trajectory 1, trajectory 4), unknown ⁇ are also generated.
- the hypothesis generation unit 32 After step S6, the hypothesis generation unit 32 generates all sets of trajectory link candidate / identification information pairs that satisfy the first condition to the third condition (step S7). A set of one trajectory link candidate / identification information pair constitutes one hypothesis. The hypothesis generation unit 32 outputs the hypothesis group to the likelihood calculation unit 33.
- the likelihood calculation control unit 330 of the likelihood calculation unit 33 determines whether there is a hypothesis for which the likelihood calculation has not yet been performed in the hypothesis group generated in step S7 (step S8). When there is a temporary calculation in which likelihood calculation processing (specifically, step S9 to step S11) has not yet been performed (that is, the determination result of step S8 is “Yes”), the likelihood calculation control unit 330 does not calculate the likelihood.
- One of the calculation hypotheses is output to the trajectory link likelihood calculation unit 331 and the identification information likelihood calculation unit 332.
- the trajectory link likelihood calculation unit 331 calculates a trajectory link likelihood for each trajectory link candidate / identification information pair belonging to the hypothesis (step S9). 6 is calculated, the disappearance time t4 of the trajectory 1 and the appearance time t6 of the trajectory 3, and the disappearance position coordinates p (90, 60) of the trajectory 3 Based on the appearance position coordinates p (130, 55), the moving speed from the disappearance of the locus 1 to the appearance of the locus 3 is calculated. An error between this moving speed and a predefined moving speed is calculated, and a trajectory link likelihood is obtained from a likelihood function using the error as a parameter. Note that the method of calculating the trajectory link likelihood is not limited to this method, and may be another method.
- the trajectory link likelihood calculation unit 331 may refer to the set of position information, detection time, and trajectory number in the past fixed time provided from the trajectory link candidate generation unit 331.
- step S9 the trajectory link likelihood calculation unit 331 calculates the trajectory link likelihood for each trajectory link candidate / identification information pair belonging to the hypothesis, and thus the trajectory link likelihood corresponding to the number of trajectory link candidates / identification information pairs belonging to the hypothesis. A likelihood is calculated.
- the identification information likelihood calculation unit 332 When the identification information likelihood calculation unit 332 inputs a hypothesis whose likelihood has not been calculated, the identification information likelihood calculation unit 332 calculates the identification information likelihood of the hypothesis (step S10). In step S10, the identification information likelihood calculating unit 332 normalizes the result of calculating and integrating the identification information likelihood for each trajectory link candidate / identification information pair belonging to the hypothesis, and uses the normalized value as the identification information likelihood. . However, the identification information likelihood calculation unit 332 does not calculate the identification information likelihood for the trajectory link candidate or the trajectory link candidate / identification information pair whose identification information is “unknown”, and other trajectory link candidate / identification information pairs. Only the identification information likelihood calculated for is integrated and normalized.
- the identification information likelihood calculation unit 332 calculates a hypothetical identification information likelihood. .
- the identification information likelihood calculation unit 332 extracts the trajectory link candidate / identification information pair from the hypothesis, and performs the following processing on the trajectory link candidate / identification information pair that does not include “unknown”.
- the identification information likelihood calculation unit 332 extracts identification information from the trajectory link candidate / identification information pair. For example, the identification information likelihood calculation unit 332 extracts ID1 from ⁇ (trajectory 1, trajectory 3), ID1 ⁇ .
- the identification information likelihood calculating unit 332 reads the detection probability of the extracted identification information from the identification information probability map. As shown in FIG. 10, it can be seen that the identification information ID1 is detected by the identification information detection device 2a at times t1 and t2. Therefore, at times t1 and t2, the identification information likelihood calculation unit 332 reads the probability value from the identification information probability map (see FIG. 8) of the identification information detection device 2a. As shown in FIG.
- the identification information likelihood calculation unit 332 reads the value “0.8” of the cell c (1, 4) from the identification information probability map shown in FIG. 8 as the probability value at time t1.
- the value “0.5” of the cell c (2, 3) is read as the probability value at the time t2.
- the identification information likelihood calculation unit 332 includes a set of position information, detection time, and trajectory number in the past fixed time provided from the trajectory link candidate generation unit 31, and a table provided from the hypothesis generation unit 32. (That is, the table after the update process in step S5) may be referred to.
- the identification information ID1 is detected by the identification information detection device 2b at times t8 and t9.
- the identification information likelihood calculation unit 332 reads the probability value from the identification information probability map (see FIG. 9) of the identification information detection device 2b.
- the moving object exists in cells c (7, 1) and c (7, 0) at times t8 and t9. Therefore, the identification information likelihood calculation unit 332 reads the value “0.7” of the cell c (7, 1) from the identification information probability map shown in FIG. 9 as the probability value at time t8. Further, the value “0.8” of the cell c (7, 0) is read as the probability value at the time t9.
- the identification information likelihood calculation unit 332 converts the detection probability of the moving object at each time when the identification information is detected into the identification information probability map corresponding to the position information corresponding to the trajectory link candidate and the identification information detecting device number. Identify based on. Note that it is not necessary to specify the detection probability of the moving object for the time when the identification information is not detected or the time when the position cannot be specified because the moving object exists outside the tracking area.
- the identification information likelihood calculation unit 332 integrates the detection probabilities of the moving objects at each time specified based on the position information corresponding to the trajectory link candidate and the identification information probability map corresponding to the identification information detection device number, This is the identification information likelihood of the target trajectory link candidate / identification information pair.
- the identification information likelihood calculation unit 332 accumulates the number of probability values used for integration for each trajectory link candidate / identification information pair, and sets the accumulated result as M.
- the identification information likelihood calculation unit 332 normalizes the integration of the probability values obtained for each trajectory link candidate / identification information pair in the hypothesis, and sets the normalized value as the identification information likelihood of the hypothesis.
- this since the identification information likelihood “0.224” is calculated for ⁇ (trajectory 1, trajectory 3), ID1 ⁇ , this may be normalized. Normalization may be performed by obtaining the Mth root of the integration result of the identification information likelihood for each trajectory link candidate / identification information pair.
- M is the cumulative number of probability values (that is, the total number) used for calculating the identification information likelihood of the trajectory link candidate / identification information pair.
- the identification information likelihood calculation unit 332 calculates the fourth power of “0.224”. Normalization for obtaining the root is performed, and the normalized result is set as the identification information likelihood of the entire hypothesis [ ⁇ (trajectory 1, trajectory 3), ID1 ⁇ , ⁇ (trajectory 2, trajectory 4), unknown ⁇ ].
- the identification information likelihood calculation unit 332 calculates the identification information likelihood of ⁇ (trajectory 2, trajectory 3, trajectory 4), ID1 ⁇ , and then performs normalization to obtain the fourth power root.
- the identification information likelihood calculation unit 332 extracts the identification information ID1.
- the identification information ID1 is detected by the identification information detection device 2a at times t1 and t2 (see FIG. 10).
- the identification information likelihood calculation unit 332 reads the probability value from the identification information probability map (see FIG. 8) of the identification information detection device 2a. As shown in FIG.
- the identification information likelihood calculation unit 332 reads the value “0.1” of the cell c (1, 0) from the identification information probability map shown in FIG. 8 as the probability value at the time t1. Further, the value “0.3” of the cell c (1, 1) is read as the probability value at the time t2.
- the identification information ID1 is detected by the identification information detection device 2b at times t8 and t9 (see FIG. 10). At times t8 and t9, the identification information likelihood calculation unit 332 reads the probability value from the identification information probability map (see FIG. 9) of the identification information detection device 2b. On the other hand, as shown in FIG. 14, for the trajectory link candidate (trajectory 2, trajectory 4), the moving body exists in the cells c (7, 1) and c (7, 0) at times t8 and t9. Therefore, the identification information likelihood calculation unit 332 reads the value “0.7” of the cell c (7, 1) from the identification information probability map shown in FIG. 9 as the probability value at the time t8. Further, the value “0.8” of the cell c (7, 0) is read as the probability value at the time t9.
- the identification information likelihood calculation unit 332 calculates the trajectory link likelihood calculated in step S9 (that is, the trajectory link likelihood for each trajectory link candidate / identification information pair in the hypothesis). Degree) and the identification information likelihood calculated in step S10 are integrated (step S11).
- the identification information likelihood calculation unit 332 integrates the trajectory link likelihood and the identification information likelihood to perform likelihood integration. That is, the integration result of the trajectory link likelihood and the identification information likelihood becomes the likelihood integration result.
- the identification information likelihood calculation unit 332 may perform weighting between the trajectory link likelihood and the identification information likelihood. For example, the trajectory link likelihood and the identification information likelihood may be multiplied by a weighting factor, and then integration of both may be performed.
- the identification information likelihood calculation unit 332 stores a set of likelihood integration results and hypotheses. For example, if the likelihood integration result of [ ⁇ (trajectory 1, trajectory 3), ID1 ⁇ , ⁇ (trajectory 2, trajectory 4), unknown ⁇ ] is X, ⁇ [ ⁇ (trajectory 1, trajectory 3), ID1 ⁇ , ⁇ (Trajectory 2, trajectory 4), unknown ⁇ ], X>.
- step S9 the number of trajectory link likelihoods corresponding to the number of trajectory link candidate / identification information pairs belonging to the hypothesis is obtained.
- a plurality of trajectory link likelihoods calculated by the trajectory link likelihood calculating unit 331 may be integrated in step S9.
- the trajectory link likelihood calculation unit 331 performs integration of a plurality of trajectory link likelihoods.
- step S11 the integration result of the trajectory link likelihood and the identification information likelihood are integrated. That is, the integration processing of the trajectory link likelihood may be performed in step S9 or may be performed collectively at the time of integration with the identification information likelihood in step S11. In any case, the likelihood integration result finally obtained in step S11 is the same.
- step S11 the likelihood calculating unit 33 repeatedly executes the loop from step S8 to step S11. If it is determined in step S8 that there is no hypothesis that has not been calculated (ie, the determination result in step S8 is “No”), the identification information likelihood calculation unit 332 integrates the trajectory link likelihood and the identification information likelihood. A temporary structure that maximizes the result is specified (step S12). That is, the identification information likelihood calculation unit 332 refers to the set of hypotheses and likelihood integration results stored in step S ⁇ b> 11 and identifies a provisional that maximizes the likelihood integration results. The identification information likelihood calculation unit 332 outputs the maximum likelihood hypothesis to the identification result output device 4.
- the identification result output device 4 When the identification result output device 4 inputs the maximum likelihood hypothesis specified in step S12, the identification result output device 4 determines which trajectory is the trajectory of which mobile object based on the trajectory link candidate / identification information pair belonging to the maximum likelihood hypothesis. To do. For example, the relationship between the trajectory and the moving body is displayed (step S13). At that time, the identification result output device 4 may specifically display the position or locus of the moving body.
- the trajectory link candidate generation unit 31 generates a trajectory link candidate obtained by connecting the trajectories detected in the past fixed time
- the hypothesis generation unit 32 is a set of a set of trajectory link candidates and identification information.
- the likelihood calculation unit 33 refers to a set of position information and identification information detected at the same time for each hypothesis, and calculates the similarity between the position information for a certain past time and the detection position based on the identification information. Calculate as likelihood. It can be said that the set of position information and identification information detected at the same time is information indicating the identity of a large number of moving bodies detected in the past certain time.
- the conventional moving object locus identification method using only the information after the tracking is resumed without using the information before the interruption occurrence, and the moving object locus identification according to the first embodiment. Compare method.
- the conventional moving body trajectory identification method as shown in FIGS. 1 to 14, when the tracking is interrupted at times t4 to t6, the identification information is newly added to the trajectories 3 and 4 detected thereafter at time t8. It is impossible to associate the trajectory with the identification information until it is detected. Further, in FIGS. 11 to 14, the trajectory 3 and the trajectory 4 are located in the same cell at times t8 and t9, respectively. In this case, it is difficult to determine whether the identification information detected at time t8 is associated with the trajectory 3 or the trajectory 4.
- the trajectories are connected using the information detected at times t1 to t9, and the detection probability of the identification information at the detection time within a predetermined past time is used.
- the likelihood of the identification information can be calculated based on the maximum likelihood hypothesis.
- the identification information likelihood calculation unit 332 calculates the integration result of the trajectory link likelihood and the identification information likelihood, and stores a set of the likelihood integration result and the hypothesis.
- the identification information likelihood calculation unit 332 estimates the hypothesis with the highest likelihood integration result as the maximum likelihood hypothesis.
- a hypothesis may be specified only from the identification information likelihood.
- the likelihood calculation unit 33 does not need to include the trajectory link likelihood calculation unit 331 and does not calculate the trajectory link likelihood (step S9). Further, since the trajectory link likelihood is not calculated, it is not necessary to integrate the trajectory link likelihood and the identification information likelihood (step S11).
- the identification information likelihood calculation unit 332 calculates the identification information likelihood of the hypothesis in step S10.
- a set of the hypothesis and the identification information likelihood may be stored.
- the identification information likelihood calculating unit 332 may specify a hypothesis having the largest identification information likelihood. Other processes are the same as those in the first embodiment.
- FIG. 15 is a block diagram showing a moving object trajectory identification system according to the second embodiment of the present invention.
- the moving body locus identification system according to the second embodiment includes a position information detection device 1, an identification information detection device 2, a moving body locus identification device 3a, and an identification result output device 4.
- the moving body trajectory identification device 3 a includes a trajectory link candidate generation unit 31, a hypothesis generation unit 32, and a map storage unit 37.
- trajectory identification device 3a comprises the likelihood calculation part 33a and the locus
- the trajectory link likelihood calculation unit 331a calculates the trajectory link likelihood for each trajectory link candidate generated by the trajectory link candidate generation unit 31.
- the method of calculating the trajectory link likelihood performed by the trajectory link likelihood calculating unit 331a is the same as the trajectory link likelihood calculating unit 331 in the first embodiment. That is, the trajectory link likelihood calculating unit 331a according to the second embodiment performs calculation processing when the trajectory link candidate includes a trajectory detected at the same time and when the trajectory detected at a different time is included. Further, the trajectory link likelihood calculating unit 331a according to the second embodiment performs a calculation process when the number of trajectories included in the trajectory link candidates is two and when there are three or more. Furthermore, when the trajectory link candidate includes only one trajectory, the trajectory link likelihood calculation unit 331a of the second embodiment sets the trajectory link likelihood of the trajectory link candidate to a constant (eg, “1”). To do.
- a constant eg, “1”.
- the trajectory link likelihood calculation unit 331a outputs only trajectory link candidates whose calculated trajectory link likelihood is equal to or greater than the threshold to the hypothesis generation unit 32. Therefore, a trajectory link candidate whose trajectory link likelihood is less than the threshold is excluded from the processing target.
- the threshold value of the trajectory link likelihood is determined in advance.
- the trajectory link likelihood calculation unit 331a may output only the trajectory link candidates whose trajectory link likelihood exceeds the threshold to the hypothesis generation unit 32.
- the hypothesis generation unit 32 generates a hypothesis using the trajectory link candidate input from the trajectory link likelihood calculation unit 331a. Other processes of the hypothesis generation unit 32 are the same as those in the first embodiment.
- the likelihood calculation unit 33a calculates the identification information likelihood for each hypothesis generated by the hypothesis generation unit 32, and sends the hypothesis having the highest identification information likelihood to the identification result output device 4.
- FIG. 16 is a block diagram illustrating a configuration of the likelihood calculating unit 33a.
- the likelihood calculation unit 33a includes a likelihood calculation control unit 330 and an identification information likelihood calculation unit 332, and does not include the trajectory link likelihood calculation unit 331 used in the first embodiment. That is, the likelihood calculation unit 33a according to the second embodiment is different from the likelihood calculation unit 33 according to the first embodiment in that the trajectory link likelihood is not calculated.
- the likelihood calculation control unit 330 outputs a hypothesis that the identification information likelihood has not been calculated to the identification information likelihood calculation unit 332. Similar to the first embodiment, the identification device likelihood calculation unit 332 calculates the identification information likelihood for the trajectory link candidate / identification information pair for each hypothesis. However, in the second embodiment, the identification information likelihood calculation unit 332 outputs the hypothesis having the highest identification information likelihood to the identification result output device 4. In the second embodiment, the identification information likelihood calculation unit 332 determines a hypothesis to be sent to the identification result output device 4 based on the identification information likelihood, not on the integration result of the trajectory link candidate and the identification information likelihood.
- the trajectory link candidate generation unit 31, trajectory link likelihood calculation unit 331a, hypothesis generation unit 32, and likelihood calculation unit 33a are, for example, mobile It is realized by a CPU of a computer that operates in accordance with a locus identification program. That is, a program storage device (not shown) of a computer stores a moving body trajectory identification program, and the CPU reads the program, so that a trajectory link candidate generation unit 31, a trajectory link likelihood calculation unit 331a, and a hypothesis generation unit. 32 and the function of the likelihood calculator 33a.
- the trajectory link candidate generation unit 31, the trajectory link likelihood calculation unit 331a, the hypothesis generation unit 32, and the likelihood calculation unit 33a may be realized by separate hardware.
- the likelihood calculation control unit 330 and the identification information likelihood calculation unit 332 may also be realized by separate hardware.
- FIGS. 17 and 18 are flowcharts illustrating processing of the moving object trajectory identification device 3a according to the second embodiment.
- the trajectory link candidate generation unit 31 acquires a set of trajectory number, detection time, and position coordinates from the position information detection device 2 (step S21).
- step S21 the trajectory link likelihood generating unit 31 inputs a set of the latest position information, detection time, and trajectory number, and additionally records them, and deletes the oldest position information to record the position information history. Update is performed (step S22). As a result, the trajectory link candidate generation unit 31 holds a plurality of pieces of position information in the past certain time from the current time.
- steps S21 and S22 in the second embodiment are the same as those in the first embodiment.
- the trajectory link candidate generation unit 31 generates a trajectory link candidate by combining the stored trajectory numbers, and outputs the trajectory link candidate to the trajectory link likelihood generation unit 331a (step S23).
- the trajectory link candidate when position coordinates of different trajectories are detected at the same time, the trajectory link candidate generation unit 31 calculates the position coordinates of the barycentric point.
- the calculation process of the locus connection candidates and the position coordinates of the barycentric points in the second embodiment is the same as step S3 in the first embodiment.
- the trajectory link candidate generation unit 31 obtains a set of position information, detection time, and trajectory number in the past fixed time (that is, a set of position information, detection time, and trajectory number obtained in the update process in step S22). Is output to the trajectory link likelihood calculation unit 331a and the likelihood calculation unit 33a. When the barycentric point of the position coordinate detected at the same time is calculated for the position coordinates of different trajectories, the barycentric point is also output to the likelihood calculating unit 33. Information input to the likelihood calculator 33 may be performed via the trajectory link likelihood calculator 331a and the hypothesis generator 32.
- the trajectory link likelihood calculating unit 331a calculates the trajectory link likelihood for each trajectory link candidate input in step S23 (step S24).
- the processing for calculating the trajectory link likelihood for each trajectory link candidate in the second embodiment is the same as the trajectory link likelihood calculating process of the trajectory link likelihood calculating unit 331 according to the first embodiment.
- the trajectory link likelihood calculation unit 331a compares the trajectory link likelihood with a predetermined threshold value, and sends only trajectory link candidates whose trajectory link likelihood is equal to or greater than the threshold value to the hypothesis generation unit 32.
- the hypothesis generation unit 32 acquires a set of detection time, identification information, and identification information detection device number from the identification information detection device 2 (step S25).
- the hypothesis generation unit 32 additionally registers the correspondence relationship between the detection time, the identification information, and the identification information detection device number in the table shown in FIG. 10, and deletes the oldest identification information (step S26).
- the hypothesis generation unit 32 not only stores the updated table content by itself, but also outputs it to the likelihood calculation unit 33a. Steps S25 and S26 in the second embodiment are the same as steps S4 and S5 in the first embodiment (see FIG. 4).
- the hypothesis generation unit 32 uses the identification information stored in the table updated in step S26 and the trajectory link candidate input from the trajectory link likelihood calculation unit 331a in step S24 to generate a trajectory link candidate / identification information pair. All combinations are generated (step S27).
- Step S27 in the second embodiment is the same as step S6 in the first embodiment except that the trajectory link candidate input from the trajectory link likelihood calculating unit 331a is used.
- the hypothesis generation unit 32 generates all sets (namely, hypotheses) of trajectory link candidate / identification information pairs that satisfy the first to third conditions (step S28).
- Step S28 in the second embodiment is the same as step S7 in the first embodiment.
- the hypothesis generation unit 32 outputs the generated hypothesis group to the likelihood calculation unit 33a.
- the likelihood calculation control unit 330 of the likelihood calculation unit 33a determines whether or not a hypothesis of actual likelihood calculation (ie, step S30 is not executed) exists in the hypothesis group input in step S28 (step S29). ). When there is a hypothesis whose likelihood has not been calculated (that is, the determination result “Yes” in step S ⁇ b> 29), the likelihood calculation control unit 330 sends one of the hypotheses whose likelihood has not been calculated to the identification information likelihood calculation unit 332. .
- the identification information likelihood calculation unit 332 calculates the identification information likelihood of the input hypothesis (step S30).
- the hypothetical identification information likelihood calculation process in the second embodiment is the same as that in the first embodiment. That is, the identification information likelihood calculation unit 332 calculates the identification information likelihood for each trajectory link candidate / identification information pair belonging to the hypothesis, and normalizes by adding the identification information likelihood of each trajectory link candidate / identification information pair. Let the value be the identification information likelihood of the hypothesis.
- the identification information likelihood calculating unit 332 stores the hypothesis and the identification information likelihood in association with each other.
- the likelihood calculating unit 33a repeatedly executes steps S29 and S30.
- the identification information likelihood calculation unit 332 determines the maximum likelihood of the identification information likelihood. Estimate as a hypothesis (step S31). The identification information likelihood calculation unit 332 sends the maximum likelihood hypothesis to the identification result output device 4.
- the identification result output device 4 determines which trajectory is the trajectory of which moving body based on the trajectory link candidate / identification information pair belonging to the maximum likelihood hypothesis and outputs the determined trajectory. (Step S32).
- the identification result output device 4 may specifically display the position or locus of the moving body. Step S32 in the second embodiment is the same as step S13 in the first embodiment (see FIG. 5).
- the trajectory link likelihood calculation unit 331a calculates the trajectory link likelihood of the trajectory link candidate and does not output the trajectory link candidate whose trajectory link likelihood is less than the threshold to the hypothesis generation unit 32.
- the trajectory link candidate is removed in advance so that it is not used for hypothesis generation. Accordingly, it is possible to reduce the workload of the process of generating the trajectory link candidate / identification information pair (step S27) and the process of generating the hypothesis (step S28) in the hypothesis generation unit 32.
- the locus and the identification information can be associated with each other at high speed. That is, it is possible to determine at high speed which locus is the locus of which moving body.
- FIG. 19 is a block diagram showing a moving object trajectory identification system according to the third embodiment of the present invention.
- the moving body locus identification system according to the third embodiment includes a position information detection device 1, an identification information detection device 2, a moving body locus identification device 3b, and an identification result output device 4.
- the moving body trajectory identification device 3b according to the third embodiment includes a trajectory link candidate generation unit 31, a hypothesis generation unit 32, and a map storage unit 37.
- the moving body trajectory identification device 3b includes a likelihood calculation unit 33b and a likelihood storage unit 34.
- the likelihood storage unit 34 stores, for each set of track number and identification information of the track being tracked, a probability that the track is associated with the identification information (hereinafter referred to as “trajectory / identification information correspondence probability”). .
- the likelihood storage unit 34 stores a trajectory / identification information correspondence probability for each set of trajectory number and identification information.
- the trajectory / identification information correspondence probability is calculated by the likelihood calculation unit 33 b and stored in the likelihood storage unit 34.
- the likelihood calculation unit 33b calculates and integrates the trajectory link likelihood and the identification information likelihood for each hypothesis. In addition, the likelihood calculating unit 33b estimates the hypothesis having the maximum integrated value of the trajectory link likelihood and the identification information likelihood as the maximum likelihood hypothesis, and sends it to the identification result output device 4.
- the likelihood calculation unit 33b calculates the identification information likelihood in the same manner as in the first embodiment, and is stored in the likelihood storage unit 34. And the integrated value is used as the identification information likelihood of the trajectory link candidate / identification information pair. Specifically, the likelihood calculating unit 33b stores the trajectory / identification information stored in the likelihood storage unit 34 for the latest trajectory (trajectory number) and identification information set for the trajectory link candidate of the trajectory link candidate / identification information pair. The correspondence probability is read out and used for the above integration process. The likelihood calculation unit 33 calculates the identification information likelihood of the hypothesis by accumulating and normalizing the identification information likelihood of the trajectory link candidate / identification information pair.
- the likelihood calculating unit 33b obtains a trajectory / identification information correspondence probability based on the identification information likelihood for each trajectory link candidate / identification information pair in each hypothesis, and stores it in the likelihood storage unit 34.
- the identification information likelihood (Q is assumed to be associated with a pair of the identification information ( ⁇ ) of the trajectory link candidate / identification information pair and the latest trajectory (trajectory number ⁇ ) in the trajectory link candidate. )
- a trajectory / identification information correspondence probability As a trajectory / identification information correspondence probability. This means that the probability that the trajectory with the trajectory number ⁇ is associated with the mobile object with the identification information ⁇ is Q.
- the trajectory / identification information correspondence probability stored in the likelihood storage unit 34 is read out next time when the likelihood calculation unit 33b calculates the identification information likelihood of the trajectory link candidate / identification information pair belonging to the hypothesis. It is.
- FIG. 20 is a block diagram illustrating a configuration of the likelihood calculating unit 33b according to the third embodiment.
- the likelihood calculation unit 33b includes a likelihood calculation control unit 330, a trajectory link likelihood calculation unit 331, and an identification information likelihood calculation unit 332b. Since the likelihood calculation control unit 330 and the trajectory link likelihood calculation unit 331 in the third embodiment are the same as those in the first embodiment, description thereof is omitted.
- the identification information likelihood calculation unit 332b calculates the identification information likelihood of each hypothesis. When calculating the identification information likelihood for each trajectory link candidate / identification information pair belonging to the hypothesis, the identification information likelihood calculation unit 332b first temporarily calculates the identification information likelihood by the same calculation method as in the first embodiment. For convenience, this identification information likelihood is referred to as temporary identification information likelihood.
- the identification information likelihood calculation unit 332 b reads the trajectory / identification information correspondence probability from the likelihood storage unit 34 for the combination of the identification information and the latest trajectory in the trajectory link candidate / identification information pair of interest. For example, when the trajectory 2 shows the latest trajectory in ⁇ (trajectory 1, trajectory 2), ID1 ⁇ , the trajectory / identification information correspondence probability is read from the likelihood storage unit 34 for “trajectory 2, ID1”.
- the identification information likelihood calculating unit 332b sets the identification information likelihood of the trajectory link candidate / identification information pair focusing on the integrated result of the temporary identification information likelihood and the trajectory / identification information correspondence probability.
- the identification information likelihood calculation unit 332b performs an integration process of the temporary identification information likelihood and the locus / identification information correspondence probability described below. That is, the identification information likelihood calculation unit 332b integrates the temporary identification information likelihood and the trajectory / identification information correspondence probability, and sets the integration result as the identification information likelihood of the trajectory link candidate / identification information pair. Alternatively, the identification information likelihood calculation unit 332b multiplies the provisional identification information likelihood and the trajectory / identification information correspondence probability by a weighting factor and adds the weighted coefficient to obtain the identification information likelihood of the trajectory link candidate / identification information pair. . Alternatively, the identification information likelihood calculation unit 332b multiplies the temporary identification information likelihood and the trajectory / identification information correspondence probability by a weighting factor and adds the result to the identification information likelihood of the trajectory link candidate / identification information pair.
- the identification information likelihood calculation unit 332b calculates the identification information likelihood of the entire hypothesis by integrating and normalizing the identification information likelihood for each trajectory link candidate / identification information pair, and integrates it with the identification information likelihood. .
- the above processing in the third embodiment is the same as that in the first embodiment.
- the identification information likelihood calculation unit 332 b estimates the hypothesis with the highest integrated value as the maximum likelihood hypothesis and sends it to the identification result output device 4.
- the identification information likelihood calculating unit 332b calculates the integrated value of the trajectory link likelihood and the identification information likelihood for all hypotheses to estimate the maximum likelihood hypothesis, and then the trajectory from the trajectory link candidate / identification information pair belonging to each hypothesis.
- the identification information correspondence probability is obtained and stored in the likelihood storage unit 34.
- the trajectory link candidate generation unit 31, the hypothesis generation unit 32, and the likelihood calculation unit 33b (including the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, and the identification information likelihood calculation unit 332b)
- the program storage unit (not shown) of the computer stores the moving body trajectory identification program, and the CPU reads the program, so that the trajectory link candidate generation unit 31, the hypothesis generation unit 32, and the likelihood calculation unit.
- the function of 33b is realized.
- the trajectory link candidate generation unit 31, the hypothesis generation unit 32, and the likelihood calculation unit 33b may be realized by separate hardware.
- the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, and the identification information likelihood calculation unit 332b may also be realized by separate hardware.
- FIG. 21 and FIG. 22 are flowcharts illustrating the processing of the moving object trajectory identification device 3b according to the third embodiment.
- symbol same as the flowchart shown in FIG.4 and FIG.5 is attached
- subjected and the description is abbreviate
- the process of generating a hypothesis group in steps S1 to S7 in FIG. 21 is the same as steps S1 to S7 in FIG. However, in step S7, a hypothesis that satisfies all of the first to third conditions described in the first embodiment may be generated.
- the hypothesis likelihood cannot be calculated because the identification device likelihood is not calculated in a hypothesis that does not satisfy the third condition, but in the third embodiment, the locus / identification information correspondence probability stored in the likelihood storage unit 34 is calculated. This is because the likelihood of the hypothesis can be calculated by referring to it.
- the hypothesis generation unit 32 outputs all the generated hypotheses to the likelihood calculation unit 33b.
- the likelihood calculation control unit 330 of the likelihood calculation unit 33b determines whether or not a hypothesis whose likelihood has not been calculated exists in the input hypothesis group (step S8). When there is a hypothesis that is not yet calculated (that is, a temporary in which steps S9 to S11 are not executed) (that is, the determination result “Yes” in step S8), the likelihood calculation control unit 330 does not calculate the likelihood.
- One of the hypotheses is output to the trajectory link likelihood calculation unit 331 and the identification information likelihood calculation unit 332b.
- the trajectory link likelihood calculation unit 331 receives a hypothesis that has not been calculated, and calculates a trajectory link likelihood for each trajectory link candidate / identification information pair belonging to the hypothesis (step S9).
- Steps S8 and S9 in the third embodiment are the same as step S9 in the first embodiment.
- the identification information likelihood calculation unit 332b calculates a provisional identification information likelihood for each trajectory link candidate / identification information pair belonging to a hypothesis whose likelihood has not been calculated (step S10a).
- the temporary identification information likelihood is calculated for each trajectory link candidate / identification information pair in step S10 (see FIG. 5) of the first embodiment. This is the same as the method of calculating.
- the further calculation performed on the calculation result in step S10a is used as the identification information likelihood of the trajectory link candidate / identification information pair. Therefore, the value calculated in step S10a is “provisional identification information likelihood”. ".
- the identification information likelihood calculation unit 332b reads the trajectory / identification information correspondence probability from the likelihood storage unit 34 for each trajectory link candidate / identification information pair.
- the trajectory / identification information correspondence probability is a probability value stored in association with the trajectory number and identification information pair of the latest trajectory in the trajectory link candidate / identification information pair.
- the temporary identification information likelihood calculated in step S10a and the trajectory / identification information correspondence probability are integrated (step S10b). Since this integration method has already been described, it is omitted here. This integration result is the identification information likelihood for each trajectory link candidate / identification information pair.
- step S10b the identification information likelihood calculation unit 332b integrates and normalizes the identification information likelihood for each trajectory link candidate / identification information pair, thereby calculating the identification information likelihood of the entire hypothesis.
- the calculation process in the third embodiment is the same as that in the first embodiment.
- the identification information likelihood calculating unit 332 integrates the trajectory link likelihood for each trajectory link candidate / identification information pair calculated in step S9 and the hypothetical identification information likelihood calculated in step S10b (step S11).
- the integration method in the third embodiment is the same as that in the first embodiment.
- the trajectory link likelihood for each hypothetical trajectory link candidate / identification information pair is integrated in step S9, and the integration result and the identification information likelihood are integrated in step S11. Also good.
- step S11 the identification information likelihood calculation unit 332b stores a set of likelihood integration results and hypotheses.
- the likelihood calculating unit 33b repeatedly executes steps S8 to S11. If it is determined in step S8 that there is no hypothesis for which the likelihood has not been calculated (that is, the determination result “No” in step S8), the identification information likelihood calculation unit 332b determines the trajectory link candidate / identification information pair in all hypotheses. Based on the above, the trajectory / identification information correspondence probability reflecting the result of the current likelihood calculation process (loop from step S8 to step S11) is stored in the likelihood storage unit 34 (step S11b).
- the identification information likelihood calculating unit 332b associates the identification information in the trajectory link candidate / identification information pair with the latest trajectory pair in the trajectory link candidate, and the identification information of the trajectory link candidate / identification information pair.
- the likelihood is stored in the likelihood storage unit 34 as a locus / identification information correspondence probability.
- a plurality of hypotheses include a trajectory link candidate / identification information pair having a common pair of the latest trajectory and identification information. For example, ⁇ (trajectory 1, trajectory 4), ID1 ⁇ is included in one hypothesis, ⁇ (trajectory 2, trajectory 4), ID1 ⁇ is included in another hypothesis, and ⁇ (trajectory 3, trajectory 3,) is included in another hypothesis.
- the locus 4 ID1 ⁇ is included and the locus 4 is the latest locus.
- the latest trajectory and identification information pair is “trajectory 4, ID1”.
- the identification information likelihood calculated for the plural trajectory link candidates / identification information pairs is added.
- the addition result may be stored as the latest locus and identification information correspondence probability regarding the identification information.
- the identification information likelihood calculation unit 332b normalizes the trajectory / identification information correspondence probability so that the total of the trajectory / identification information correspondence probability becomes “1” for each trajectory, and the normalization result is used as the likelihood.
- FIG. For example, for the two identification information ID1 and ID, the identification information likelihood of the set of the locus 4 and ID1 is X, and the identification information likelihood of the combination of the locus 4 and ID2 is Y. At this time, X and Y are not used as the trajectory / identification information correspondence probability as they are, but the trajectory / identification information correspondence probability of the trajectory 4 and ID1 pair and the trajectory / identification information correspondence probability of the trajectory 4 and ID2 pair. X and Y are normalized so that the sum is “1”.
- the normalized trajectory / identification information correspondence probabilities are x and y, respectively, the identification information likelihood calculation unit 332b performs the following calculation.
- the identification information likelihood calculation unit 332b performs normalization by multiplying the individual identification information likelihoods X and Y before normalization by the reciprocal “1 / (X + Y)” of the sum.
- step S11b only the process of storing the trajectory / identification information correspondence probability in the likelihood storage unit 34 may be performed.
- the process of storing the trajectory / identification information correspondence probability in the likelihood storage unit 34 is performed, and the trajectory of the trajectory link candidate / identification information pair among the trajectory / identification information correspondence probabilities stored in the likelihood storage unit 34.
- step S11b the trajectory / identification information correspondence probability stored in the likelihood storage unit 34 is then used when executing the loop of steps S8 to S11.
- the identification information likelihood calculation unit 332b estimates the maximum likelihood hypothesis that maximizes the likelihood integrated value calculated in step S11, and sends it to the identification result output device 4 (step S12). Based on the maximum likelihood hypothesis, the identification result output device 4 determines and outputs which trajectory is the trajectory of which moving body (step S13). Steps S12 and S13 in the third embodiment are the same as those in the first embodiment.
- Example 3 has the same effect as Example 1. Furthermore, in the third embodiment, since the likelihood is calculated considering not only the information of the past fixed time but also the previous time from the current time, the correspondence between the trajectory and the identification information is estimated with higher accuracy. can do.
- FIG. 23 is a schematic diagram for explaining the effect of the third embodiment.
- processing for calculating likelihood for each hypothesis at time u that is, a loop from step S8 to step S11
- data in a time zone U that goes back a certain time from time u to the past is used.
- the trajectory / identification information correspondence probability obtained by executing the loop of steps S8 to S11 for each hypothesis reflects the data of the time zone U (more precisely, the previous time).
- the likelihood integration result in step S11 is calculated in consideration of the locus / identification information correspondence probability calculated at time u.
- This likelihood integration result reflects not only the data of time zone V but also the data of time zone U before that.
- the likelihood for each hypothesis is calculated in consideration of past information as described above, the correspondence between the trajectory and the identification information can be determined with higher accuracy.
- FIG. 24 is a block diagram showing a moving body trajectory identification system according to Embodiment 4 of the present invention.
- the moving body locus identification system according to the fourth embodiment includes a position information detection device 1, an identification information detection device 2, a moving body locus identification device 3c, and an identification result output device 4.
- the moving body trajectory identification device 3 according to the fourth embodiment includes an information history change determination unit 35, a trajectory link candidate generation unit 31c, a hypothesis generation unit 32c, a likelihood calculation unit 33c, and a map storage unit 37.
- the information history change determination unit 35 acquires a set of a trajectory number, a detection time, and a position coordinate from the position information detection device 1, and sets a detection time, identification information, and an identification information detection device number from the identification information detection device 2. To get.
- the information history change determination unit 35 holds the position coordinates, the detection time, and the trajectory number acquired from the position information detection apparatus 1 in a certain past time from the current time.
- the information history change determination unit 35 holds the correspondence relationship between the identification information acquired from the identification information detection device 2 in the past certain time from the current time, the detection time, and the identification information detection device number. At that time, the data is stored in the table format shown in FIG.
- the information history change determination unit 35 holds the past fixed time among the position information / identification information acquired from the position information detection device 1 and the identification information detection device 2, so that the position information /
- the retained information is updated by deleting the position information / identification information of the oldest time. For example, when the latest position information, detection time, and trajectory number are acquired from the position information detection apparatus 1, they are stored and the position information at the oldest time is deleted. Similarly, when the latest identification information, the detection time, and the identification information detection device number are acquired from the identification information detection device 2, it is additionally registered in the table shown in FIG. 10, and the identification information at the oldest time is deleted from the table. .
- the update process by the information history change determination unit 35 is the same as steps S1 and S2 of the trajectory link candidate generation unit 31 and steps S4 and S5 of the hypothesis generation unit 32 in the first embodiment.
- the information history change determination unit 35 outputs the held position information / identification information to the likelihood calculation unit 33c. That is, the likelihood calculation unit 33c inputs a set of detection time, position coordinates, and trajectory number, and the table contents of FIG. 10 (that is, updated identification information).
- the information history change determination unit 35 When the information history change determination unit 35 newly acquires position information / identification information from the position information detection device 1 and the identification information detection device 2 and performs an update, is there a change in the trajectory number and the identification information before and after the update? Determine whether or not. If neither the trajectory number nor the identification information is changed, the same hypothesis as the previous one can be obtained even if the hypothesis is generated again. In this case, the information history change determination unit 35 instructs the likelihood calculation unit 33c to estimate the maximum likelihood hypothesis based on the hypothesis used for the previous likelihood calculation. In this case, no information is input to the trajectory link candidate generation unit 31c and the hypothesis generation unit 32c.
- the same maximum likelihood hypothesis as the previous one is estimated when the position coordinates change after the update or the identification information detecting device number associated with the identification information changes. Not necessarily.
- the information history change determination unit 35 provides a set of updated position coordinates, detection time, and track number to the track connection candidate generation unit 31c when a change occurs in the track number or the identification information.
- the table content indicating the correspondence between the identification information, the detection time, and the identification information detection device number is provided to the hypothesis generation unit 32c.
- the trajectory link candidate generation unit 31c generates a trajectory link candidate based on the position information / identification information supplied from the information history change determination unit 35.
- the locus connection candidate generation unit 31c also calculates the position coordinates of the barycentric point. Then, the trajectory link candidate generation unit 31c outputs the trajectory link candidate to the hypothesis generation unit 32c.
- the position coordinates of the center of gravity are calculated, the position coordinates are also output to the hypothesis generation unit 32c.
- the hypothesis generation unit 32c generates all hypotheses based on the trajectory link candidate input from the trajectory link candidate generation unit 31c and the updated table content input from the information history change determination unit 35, and the likelihood calculation unit To 33c.
- trajectory link candidate generation unit 31c and the hypothesis generation unit 32c according to the fourth embodiment do not perform updating so as to hold the position information / identification information for a certain period of time in the past, and the trajectory link candidate according to the first embodiment. Different from the generation unit 31 and the hypothesis generation unit 32.
- the likelihood calculating unit 33c receives an instruction from the information history change determining unit 35 to estimate the maximum likelihood hypothesis based on the hypothesis group used in the previous likelihood calculation, the trajectory link likelihood and the identification information likelihood for each hypothesis. The degree is calculated and the integrated value is calculated. The likelihood calculation unit 33c estimates the hypothesis having the highest integrated value of the trajectory link candidate and the identification information likelihood as the maximum likelihood hypothesis, and transmits the hypothesis to the identification result output device 4.
- the likelihood calculation unit 33c calculates a trajectory link likelihood and an identification information likelihood for each hypothesis, and calculates an integrated value thereof.
- the likelihood calculating unit 33 c estimates the hypothesis having the highest integrated value of the trajectory link likelihood and the identification information likelihood as the maximum likelihood hypothesis, and sends it to the identification result output device 4.
- the likelihood calculation unit 33c inputs a set of detection time, position coordinates, and trajectory number from the information history change determination unit 35, and updated table contents (see FIG. 10). The likelihood calculation unit 33c uses the information updated by the information history change determination unit 35 when calculating the trajectory link likelihood and the identification information likelihood.
- FIG. 25 is a block diagram illustrating a configuration of a likelihood calculating unit 33c according to the fourth embodiment.
- the likelihood calculation unit 33c includes a likelihood calculation control unit 330, a trajectory link likelihood calculation unit 331, an identification information likelihood calculation unit 332c, and a hypothesis storage unit 333.
- the hypothesis storage unit 333 is a memory that stores a hypothesis group whose maximum likelihood hypothesis has already been estimated.
- the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, and the identification information likelihood calculation unit 332c according to the fourth embodiment are the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, and the identification according to the first embodiment.
- the same processing as the information likelihood calculation unit 332 is performed.
- a likelihood calculation control unit 330, a trajectory link likelihood calculation unit 331, and an identification information likelihood calculation unit 332c Uses a hypothesis group stored in the hypothesis storage unit 333 as a processing target.
- the hypothesis group is set as a processing target.
- the identification information likelihood calculation unit 332c receives a command for estimating the maximum likelihood hypothesis from the previous hypothesis group, and in the trajectory link candidate included in the hypothesis group, at different position coordinates detected at the same time. If there is a position coordinate whose centroid point has not been calculated, the coordinate of the centroid point is calculated.
- Information history change determination unit 35, trajectory link candidate generation unit 31c, hypothesis generation unit 32c, and likelihood calculation unit 33c is realized by, for example, a CPU of a computer that operates according to the moving object locus identification program.
- the program storage unit (not shown) of the computer stores the moving body trajectory identification program, and the CPU reads the program, so that the information history change determination unit 35, the trajectory link candidate generation unit 31c, and the hypothesis generation unit. 32c and the function of the likelihood calculator 33c are realized.
- the information history change determination unit 35, the trajectory link candidate generation unit 31c, the hypothesis generation unit 32c, and the likelihood calculation unit 33c may be configured by separate hardware.
- the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332c, and the hypothesis storage unit 333 may also be configured by separate hardware.
- the hypothesis storage unit 333 may be arranged outside the likelihood calculation unit 33c.
- FIG.26 and FIG.27 is a flowchart which shows the process of the mobile body locus
- processing similar to that in the first embodiment is denoted by the same reference numeral, and description thereof is omitted.
- the information history change determination unit 35 acquires a set of a trajectory number, a detection time, and position information from the position information detection device 1 (step S41).
- the information history change determination unit 35 inputs and stores the latest set of position information, detection time, and trajectory number, and updates the position information history by deleting the oldest position information (step S42). .
- Steps S41 and S42 in the fourth embodiment are the same as steps S1 and S2 in the first embodiment. However, in Example 4, the information history change determination unit 35 outputs the set of the updated position information, detection time, and trajectory number to the likelihood calculation unit 33c in step S42.
- the information history change determination unit 35 acquires a set of identification information detection time, identification information, and identification information detection device number from the identification information detection device 2 (step S43).
- the information history change determination unit 35 acquires a set of the latest detection time, identification information, and identification information detection device number, additionally registers it in the table shown in FIG. 10, and deletes the oldest identification information (step S44).
- Steps S43 and S44 in the fourth embodiment are the same as steps S4 and S5 in the first embodiment. However, the information history change determination unit 35 outputs the updated table content to the likelihood calculation unit 33c in step S44.
- the information history change determination unit 35 determines whether or not the trajectory number has changed before and after the update in step S42, and whether or not the identification information registered in the table after the update in step S44 has changed (step). S45).
- the change of the trajectory number / identification information occurs due to the addition or discard of the trajectory number / identification information.
- step S45 when a change has occurred in at least one of the trajectory number and the identification information (that is, the determination result “Yes”), the information history change determination unit 35 determines the updated position information, detection time, and trajectory.
- the number set is output to the trajectory link candidate generation unit 31c, and the updated table contents (indicating the relationship between the detection time, the identification information, and the identification information detection device number) are output to the hypothesis generation unit 32c.
- the trajectory link candidate generation unit 31c generates all trajectory link candidates based on the set of position information, detection time, and trajectory number, and outputs them to the hypothesis generation unit 32c (step S3). In addition, when position coordinates of different trajectories are detected in the generated trajectory link candidate at the same time, the trajectory link candidate generation unit 31c also calculates the position coordinates of the barycentric point and outputs the position coordinates to the likelihood calculation unit 33c. To do.
- the hypothesis generation unit 32c uses the identification information stored in the updated table input from the information history change determination unit 35 and the trajectory link candidate input from the trajectory link candidate generation unit 31c in step S3. All the identification information pairs are generated (step S6). Next, the hypothesis generation unit 32c generates all hypotheses that satisfy the first to third conditions described in the first embodiment (step S7). The hypothesis generation unit 32c outputs the hypothesis group generated in step S7 to the likelihood calculation unit 33c. Steps S3, S6, and S7 in the fourth embodiment are the same as those in the first embodiment.
- the likelihood calculation unit 33c executes the same steps S8 to S12 as those in the first embodiment with the hypothesis group as a processing target. More specifically, the likelihood calculation control unit 330 determines whether there is a hypothesis whose likelihood has not been calculated in addition to the hypothesis group input in step S7 (step S8). When there is a hypothesis whose likelihood has not been calculated (that is, the determination result “Yes” in step S8), the likelihood calculation control unit 330 assigns one of the hypotheses whose likelihood has not been calculated to the trajectory link likelihood calculation unit 331 and It outputs to the identification information likelihood calculation part 332c.
- the trajectory link likelihood calculating unit 331 calculates the trajectory link likelihood for each trajectory link candidate / identification information pair belonging to the input hypothesis (step S9).
- the position coordinate and the detection time used for calculating the trajectory link likelihood are provided from the information history change determination unit 35 as a set of the updated position coordinate, the detection time, and the trajectory number.
- the identification information likelihood calculation unit 332c calculates the identification information likelihood for the input hypothesis (step S10).
- the position coordinates and the detection time are provided from the information history change determination unit 35.
- updated table contents indicating the correspondence relationship between the identification information, the detection time, and the identification information detection device number are also provided from the information history change determination unit 35. Therefore, the identification information likelihood calculation unit 332c according to the fourth embodiment can calculate the identification information likelihood of the hypothesis using the position information and the identification information, as in the first embodiment.
- the identification information likelihood calculation unit 332c integrates the hypothesis trajectory link likelihood calculated in step S9 and the hypothesis identification information likelihood calculated in step S10, and associates the integration result with the hypothesis.
- step S11 the loop from step S8 to step S11 is repeated.
- the identification information likelihood calculation unit 332c estimates the hypothesis that maximizes the integrated value of the trajectory link likelihood calculated in step 11 and the identification information likelihood as the maximum likelihood hypothesis, and performs identification. The result is output to the output device 4 (step S12).
- the identification information likelihood calculation unit 332c according to the fourth embodiment stores all hypotheses (that is, the hypothesis group generated by the hypothesis generation unit 32c) associated with the integration result of the trajectory link likelihood and the identification information likelihood as a hypothesis storage. The data is stored in the unit 333 and the stored content is updated.
- the information history change determination unit 35 performs the previous likelihood calculation on the likelihood calculation unit 33c.
- An instruction is given to estimate the maximum likelihood hypothesis based on the used hypothesis (that is, the hypothesis group stored in the hypothesis storage unit 333).
- the likelihood calculation control unit 330 acquires a hypothesis group stored in the hypothesis storage unit 333 in response to this command (step S46).
- Likelihood calculation unit 33c executes steps S8 to S12 in the same manner as in the first embodiment for the hypothesis group. First, the likelihood calculation control unit 330 determines whether there is a hypothesis whose likelihood has not been calculated among the hypotheses read from the hypothesis storage unit 333 (step S8).
- the likelihood calculation control unit 330 assigns one of the hypotheses whose likelihood has not been calculated to the trajectory link likelihood calculation unit 331 and It outputs to the identification information likelihood calculation part 332c.
- the trajectory link likelihood calculation unit 331 calculates the trajectory link likelihood for each trajectory link candidate / identification information pair belonging to the input hypothesis (step S9).
- the position coordinates and the detection time used for calculating the trajectory link likelihood are provided from the information history change determination unit 35 as a set of the updated position coordinates, the detection time, and the trajectory number.
- the identification information likelihood calculation unit 332c calculates the identification information likelihood for the input hypothesis (step S10). As described above, since the updated identification information is provided from the information history change determination unit 35, the identification information likelihood of the hypothesis can be calculated as in the first embodiment. However, for the newly added registered detection time, if there are position coordinates of different trajectories, and there are position coordinates detected at the same time, the coordinates of the centroid point are not calculated. The information likelihood calculation unit 332c calculates the position coordinates of the barycentric point and provides the identification information likelihood.
- the identification information likelihood calculation unit 332c integrates the trajectory link likelihood calculated in step S9 and the identification information likelihood calculated in step S10de, and stores the integration result and the hypothesis in association with each other (step S11). .
- the loop from step S8 to step S11 is repeated.
- the identification information likelihood calculation unit 332c estimates the hypothesis that maximizes the integrated value of the trajectory link likelihood and the identification information likelihood calculated in step S11 as the maximum likelihood hypothesis, The result is sent to the identification result output device 4 (step S12).
- Steps S11 and S12 are the same as when the hypothesis generation unit 32c newly generates a hypothesis group. However, when the hypothesis group stored in the hypothesis storage unit 333 is a processing target, the identification information likelihood calculation unit 332 c does not update the storage content of the hypothesis storage unit 333.
- the identification result output device 4 determines which trajectory is the trajectory of which moving body based on the trajectory link candidate / identification information pair belonging to the maximum likelihood hypothesis estimated in step S12 and outputs the trajectory (step S13). At this time, the identification result output device 4 may specifically display the position or locus of the moving body. Step S13 in the fourth embodiment is the same as that in the first embodiment.
- the maximum likelihood hypothesis based on the hypothesis group used in the previous likelihood calculation Is estimated.
- the same trajectory link candidate and hypothesis as in the previous time are calculated. That is, the duplication process in which the hypothesis generation unit 32c generates the same hypothesis group again is eliminated, thereby reducing the amount of calculation in the determination of the association between the locus and the identification information.
- the moving body trajectory identification device 3c instead of using the inputs from the position information detection device 1 and the identification information detection device 2 as triggers, the moving body trajectory identification device 3c starts processing from step S41 onward using an external command as a trigger. It may be. In this case, if there is no new input from the position information detection device 1 and the identification information detection device 2 after the previous estimation of the maximum likelihood hypothesis, the information is not updated in steps S42 and S44. For this reason, the trajectory number and the identification information do not change (that is, the determination result “No” in step S45). As a result, the flow proceeds to step S46. After the previous maximum likelihood hypothesis is estimated, if there is no input from the position information detection device 1 and the identification information detection device 2, the same maximum likelihood hypothesis as the previous one is estimated.
- the Rukoto Therefore, if there is no input from the position information detection device 1 and the identification information detection device 2 after the previous estimation of the maximum likelihood hypothesis, and the mobile body locus identification device 3c is instructed to start processing, the mobile object The trajectory identification device 3c may send the previously estimated maximum likelihood hypothesis to the identification result output device 4 as it is.
- the likelihood calculation units 33b and 33c are not provided with the trajectory link likelihood calculation unit 331 so that the trajectory link likelihood is not calculated. It may be.
- the identification information likelihood calculation units 332b and 332c estimate the hypothesis having the largest identification information likelihood as the maximum likelihood hypothesis without integrating the trajectory link likelihood and the identification information likelihood.
- Example 4 when the likelihood calculation unit 33c does not include the trajectory link likelihood calculation unit 331 and does not calculate the trajectory link likelihood, the information history change determination unit 35 causes a change in the identification information in step S45. If there is a change (determination result “Yes” in step S45), the flow proceeds to step S3. On the other hand, if there is a change (determination result “No” in step S45), the flow proceeds to step S3. The process proceeds to S46.
- the likelihood calculation units 33b and 33c do not include the trajectory link likelihood calculation unit 331, and the trajectory link likelihoods are included in the moving object trajectory identification devices 3b and 3c as in the second embodiment.
- a degree calculator 331a may be provided.
- the identification information likelihood calculation units 332b and 332c estimate the hypothesis having the highest identification information likelihood as the maximum likelihood hypothesis without integrating the trajectory link likelihood and the identification information likelihood. According to this configuration, the same effect as in the second embodiment can be obtained. Furthermore, Example 4 and Example 3 may be combined.
- FIG. 28 is a block diagram showing a moving body trajectory identification system according to the fifth embodiment of the present invention.
- the moving body locus identification system according to the fifth embodiment includes a position information detection device 1, an identification information detection device 2, a moving body locus identification device 3d, an identification result output device 4, and an attribute information detection device 5.
- the moving body trajectory identification device 3d according to the fifth embodiment includes a trajectory link candidate generation unit 31, a hypothesis generation unit 32d, a likelihood calculation unit 33d, and a map storage unit 37.
- the number of attribute information detection devices 5 is not limited to one, and a plurality of attribute information detection devices 5 may be provided.
- an attribute information detection device ID for uniquely specifying the attribute information detection device 5 is assigned in advance.
- the attribute information detection device ID is represented by a number and is referred to as an attribute information detection device number.
- the attribute information detection device 5 acquires the attribute information of the moving object in the tracking area 50 (see FIG. 1).
- the attribute information represents the attributes and characteristics of the moving object, and for example, the color, shape, size, weight, temperature, etc. of the moving object can be used.
- the sex and age of the person may be used as attribute information.
- the attribute information only needs to represent the attribute of the moving body, and may be information (for example, color, shape, etc.) that can be determined from an image of the moving body. Or the information (for example, weight etc.) detectable with a sensor or a measuring instrument, without imaging a moving body may be sufficient.
- the attribute information detection device 5 may be any device that can detect the attribute information of the moving object in the tracking area 50 and specify the detection time and position coordinates.
- the attribute information detection device 5 is realized by a camera, a pressure sensor, a temperature sensor, and a thermography.
- the detection areas of the attribute information detection devices 5 may be installed so as to overlap each other. Or you may install so that a detection area may not mutually overlap. Moreover, you may use together the some attribute information detection apparatus 5 from which a kind differs like a camera or a pressure sensor.
- the attribute information detection device 5 does not always detect the attribute information, and whether or not the attribute information is detected depends on the installation status of the attribute information detection device 5. For example, when the attribute information detection device 5 is installed in the tracking area, it is easy to detect the attribute information if there is a moving object near the attribute information detection device 5, but at a location away from the attribute information detection device 5. If a moving object exists, it becomes difficult to detect attribute information.
- the attribute information detection device 5 receives the attribute information of the detected moving object, the attribute information type that uniquely indicates the type of the attribute information, the detection time of the attribute information, the position coordinates, and the attribute information detection device number. Output to. If the attribute information detection device 5 tries to detect the attribute information of the moving body and fails to detect it, the attribute information, the attribute information type, and the position coordinate are set to “none”, and the detection time and the attribute information detection device number At the same time, it may be output to the moving body trajectory identification device 3d. Alternatively, even if the attribute information detection device 5 does not input any attribute information to the moving body trajectory identification device 3d, it is determined that no attribute information is detected at that time for the moving body trajectory identification device 3d. Good.
- the attribute information detection device 5 determines the position coordinates of the installation location. As the position coordinates of the moving object, is sent to the moving object locus identifying device 3d.
- An example of the attribute information detection device 5 is a pressure sensor. In this case, the attribute information detection device 5 may hold the position coordinates of its own installation position in the tracking area 50 in advance.
- the attribute information detection device 5 determines attribute information from an image captured using a camera or the like, the range of the captured image is wide, and the movement at the time when the installation information and the moving body are captured and the attribute information is extracted. It does not necessarily match the body position.
- the attribute information detection device 5 stores in advance a conversion table for converting an arbitrary position in the captured image into the position coordinates of the tracking area 50, and the position of the moving body in the image is stored in the tracking area 50 according to the conversion table. It may be converted into the position coordinates.
- a conversion equation for converting an arbitrary position coordinate in the captured image into a position coordinate in the tracking area 50 is set in advance, and the position of the moving object in the image is substituted into the conversion equation. You may convert into the position coordinate in the tracking area
- the position coordinates of the moving object at the time of detecting the attribute information are obtained by a method according to the type of the attribute information detecting device 5 (that is, the attribute information type).
- the position information by the position information detection device 1, the identification device by the identification device detection device 2, and the attribute information by the attribute information detection device 5 need to be synchronized.
- the position information detection device 1 detects the position information of the moving body
- the identification information detection device 2 detects the identification information
- the attribute information detection device 5 detects the attribute information.
- the moving object locus identification device 3d detects the position information, the identification information, and the attribute.
- the information may be buffered for a certain period of time, and the position information, identification information, and attribute information stored in the buffer may be read every certain period of time.
- the time information is not synchronized among the position information detection device 1, the identification information detection device 3, and the attribute information detection device 5, the position information, identification information, and The same detection time may be applied to the attribute information.
- the hypothesis generation unit 32d inputs the trajectory link candidate from the trajectory link candidate generation unit 31 and the identification information from the identification information detection device 2. Further, the hypothesis generation unit 32d receives the attribute information, the attribute information type, the detection time, the position coordinates, and the attribute information detection device number from the attribute information detection device 5, and holds the corresponding relationship.
- the hypothesis generation unit 32d generates all hypotheses based on the trajectory link candidate by the trajectory link candidate generation unit 31, the identification information by the identification information detection device 2, and the attribute information by the attribute information detection device 5, and the hypothesis group is estimated as the likelihood. This is output to the degree calculator 33d.
- the hypothesis generation unit 32d uses the trajectory link candidate input from the trajectory link candidate generation unit 31 and the attribute information detected by the attribute information detection device 5 in the time zone from the current time to a time that is a certain time past in the past. A pair of trajectory link candidates and attribute information is generated. Hereinafter, it is described as a trajectory link candidate with attribute information.
- the hypothesis generation unit 32d selects a trajectory link candidate that associates attribute information with the following procedure, and generates a trajectory link candidate with attribute information.
- the hypothesis generation unit 32d selects one of the attribute information input from the attribute information detection device 5, and calculates the Euclidean distance between the position coordinates of the trajectory link candidate and the position coordinates of the selected attribute information at the detection time. Calculation is made for each set of all trajectory link candidates and selected attribute information.
- the hypothesis generation unit 32d specifies a pair having the shortest Euclidean distance from the pair of trajectory link candidates and attribute information, and stores the trajectory link candidate and attribute information in the set in association with each other. This process is performed once for all the input attribute information.
- the hypothesis generation unit 32d selects the trajectory included in the trajectory link candidate. Based on the disappearance position coordinates and the disappearance time, and the appearance position coordinates and the appearance time, the discontinuous locus is interpolated to obtain the position coordinates at the detection time of the attribute information, and thus the Euclidean distance from the position coordinates of the attribute information. Is calculated.
- the trajectory link candidate includes a plurality of trajectories detected at the detection time of the attribute information, and there are a plurality of position coordinates of the moving object at the detection time, the center of gravity point and the position coordinates of the attribute information
- the Euclidean distance of is calculated.
- trajectory link candidate that is not associated with attribute information.
- trajectory link candidate associated with a plurality of attribute information.
- the trajectory link candidate generation unit 31 generates trajectory link candidates T1, T2,..., Tn, and the attribute information detection device 5 detects attribute information A1, A2,.
- the fact that there is no attribute information associated with the trajectory link candidate is represented by a character string “no attribute”.
- the hypothesis generation unit 32d for example, sets (T1, A1), (T2, A2),..., (Tn, Am), (T1, no attributes), (T2, A1), (T T3, A2, A3),..., (Tn, Am-2, Am-1, Am) are generated and held as trajectory link candidates with attribute information.
- the hypothesis generation unit 32d may generate a trajectory link candidate with attribute information by associating all input attribute information with any trajectory link candidate. Further, the hypothesis generation unit 32d, for a trajectory link candidate determined to have the smallest Euclidean distance from the position coordinates of the attribute information, if the Euclidean distance exceeds a predetermined threshold, You may determine not matching with a connection candidate.
- the hypothesis generation unit 32d uses the attribute information-added trajectory link candidate generated according to the above procedure and the identification information detected by the identification information detection apparatus 2 in the time zone from the current time to a time that is a certain time in the past. All sets of trajectory link candidates with information and identification information are generated.
- the hypothesis generation unit 32d pays attention to the trajectory link candidate with attribute information, regards it as a simple trajectory link candidate, and uses the same method as the generation method of the trajectory link candidate / identification information pair in the first embodiment to track with attribute information. All combinations of connection candidates and identification information may be generated. Hereinafter, this is referred to as a trajectory link candidate / identification information pair with attribute information.
- the hypothesis generation unit 32d extracts all sets of trajectory link candidate / identification information pairs with attribute information that satisfy the first condition to the third condition. Assume a set.
- the likelihood calculation unit 33d calculates the trajectory link likelihood, the identification information likelihood, and the attribute information likelihood for each hypothesis generated by the hypothesis generation unit 32d, and integrates them. Then, the hypothesis having the largest integrated value of the trajectory link likelihood, the identification information likelihood, and the attribute information likelihood is estimated as the maximum likelihood hypothesis and is transmitted to the identification result output device 4.
- the attribute information likelihood is the likelihood that individual attribute information associated with the trajectory link candidate in the trajectory link candidate with attribute information included in the hypothesis represents the attribute of the same mobile object. For example, when the attribute information associated with the trajectory link candidate represents a clothing color, the similarity of the clothing color may be indicated as the attribute information likelihood. Alternatively, an element other than the color of clothes, for example, careful similarity, weight similarity, age similarity, and gender similarity may be displayed according to the attribute information type.
- FIG. 29 is a block diagram illustrating a configuration of a likelihood calculating unit 33d according to the fifth embodiment.
- the likelihood calculator 33d includes a likelihood calculation controller 330d, a trajectory link likelihood calculator 331, an identification information likelihood calculator 332, and an attribute information likelihood calculator 334.
- the trajectory link likelihood calculation unit 331 and the identification information likelihood calculation unit 332 are the same as those in the first embodiment, description thereof is omitted.
- the likelihood calculation control unit 330d outputs a hypothesis whose likelihood has not been calculated to the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the attribute information likelihood calculation unit 334.
- the attribute information likelihood calculation unit 334 receives the hypothesis generated by the hypothesis generation unit 32d and calculates the attribute information likelihood for the hypothesis. When calculating the attribute information likelihood of each hypothesis, the attribute information likelihood calculating unit 334 calculates the attribute information likelihood for each trajectory link candidate included in the hypothesis, and for each trajectory link candidate belonging to the same hypothesis. The calculated attribute information likelihood is integrated and normalized to obtain the attribute information likelihood of the hypothesis.
- the attribute information likelihood calculation unit 334 presets an attribute information likelihood calculation method for each attribute information type, and calculates the optimum attribute information likelihood according to the type of attribute information included in each hypothesis. Determine how. Then, the attribute information likelihood calculation unit 334 calculates the attribute information likelihood by a calculation method according to the attribute information type.
- one type of attribute information may be associated with the trajectory link candidate, or multiple types of attribute information may be associated.
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood for the trajectory link candidate with attribute information.
- a calculation process when two attribute information items are associated with a trajectory link candidate and the attribute information types are the same will be described.
- attribute information is represented by a histogram, and the number of bins in each histogram is “m”.
- the two pieces of attribute information (histograms) are h1 and h2, and the frequencies in the i-th bin in each histogram are h1 (i) and h2 (i).
- a color of clothes can be given as an attribute information type that can be represented by such a histogram.
- the color of clothes can be represented by a color space histogram.
- the attribute information detection device 5 creates a histogram in which the number of pixels included in an image obtained by capturing a moving object is counted for each color type, and the histogram is used as attribute information.
- the attribute information likelihood calculating unit 334 calculates the Bhattacharya distance D according to Equation 4, and the attribute information likelihood is higher as the Bhattacharya distance D is shorter and the attribute information likelihood is longer as the Bhattacharya distance D is longer.
- the attribute information likelihood is determined so as to be low.
- the attribute information likelihood represented by the histogram is determined based on the Bhattacharya distance, but the attribute information likelihood may be calculated by another method.
- attribute information likelihoods corresponding to combinations of attribute information associated with trajectory link candidates are defined and held in the attribute information likelihood calculation unit 334. At this time, it is defined that the attribute information likelihood is increased for combinations with the same attribute information, and the attribute information likelihood is decreased for combinations with different attribute information.
- the attribute information associated with the trajectory link candidate is “male / male” or “female / female”, the attribute information has a high likelihood of being related to the same mobile object. Define high.
- the attribute information likelihood is defined to be low when the attribute information associated with the trajectory link candidate is “male / female”.
- the attribute information likelihood calculation unit 334 may hold the attribute information defined as described above and read the attribute information likelihood corresponding to the attribute information pattern associated with the trajectory link candidate.
- the attribute information likelihood calculation unit 334 calculates the variance of the two numerical values associated with the attribute information likelihood of the trajectory link candidate.
- the attribute information likelihood calculation unit 334 holds in advance a function (hereinafter referred to as “likelihood function”) in which the attribute information likelihood is higher as the variance value is smaller and the attribute information likelihood is lower as the variance value is larger. To do.
- the attribute information likelihood calculating unit 334 inputs the calculated variance value to the likelihood function and calculates the attribute information likelihood.
- the attribute information likelihood is calculated using variance, but the attribute information likelihood may be calculated by another method.
- the attribute information likelihood regarding the trajectory link candidate is calculated by the following procedure.
- the attribute information likelihood calculation unit 334 orders the attribute information associated with the trajectory link candidate in the order from the earliest detection time.
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood for the attribute information whose order is adjacent. For example, the likelihood is calculated for the first and second attribute information, similarly, the likelihood is calculated for the second and third attribute information, and the likelihood is calculated for the third and fourth attribute information. To do. Thereafter, the attribute information likelihood is sequentially calculated. At this time, the likelihood calculation for the two pieces of attribute information that are adjacent in order can be calculated by the calculation method in which the two pieces of attribute information are associated with the trajectory link candidate.
- the attribute information likelihood calculation unit 334 calculates the likelihood for each pair of attribute information items that are adjacent in order, then integrates the attribute information likelihoods, and a power root of the number of attribute information likelihoods integrated. Calculate By calculating the power root, the attribute information on the trajectory link candidate with attribute information is normalized.
- the attribute information likelihoods may be weighted according to the attribute information detection device number corresponding to the original attribute information and then accumulated. This is a measure that takes into account the characteristics of the attribute information detection device 5. For example, in the attribute information detection device 5 that can capture the face of a moving object (here, a person) from the front, attribute information such as gender and age is used. Can be detected with high accuracy. On the other hand, in the attribute information detection apparatus 5 that can only capture a human face from an oblique direction, there is a possibility that the detected attribute information includes an error. In this way, the integration is performed after weighting each attribute information likelihood, reflecting the detection accuracy of each attribute information detection device. Further, when weighting according to the attribute information detection device 5 is not performed, it is not necessary to output the attribute information detection device number.
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood related to the trajectory link candidate according to the following procedure.
- the attribute information likelihood calculation unit 334 orders the attribute information associated with the trajectory link candidate in the order from the earliest detection time. Then, the attribute information likelihood calculation unit 334 classifies the attribute information and calculates the attribute information likelihood for each attribute information type. When calculating the attribute information likelihood with respect to one type of attribute information type, the attribute information likelihood calculating unit 334 generates a combination of attribute information in close order with respect to the attribute information type of interest, and attribute information for each combination. Calculate the likelihood.
- the attribute information in which the order is close means that the attribute information of the attribute information type of interest is close in order, and there is another attribute information related to the same type between them. It means not. Note that there may be another attribute information related to another type between the attribute information related to the same type.
- the calculation of the attribute information likelihood for the combination of the attribute information related to the same type in which the ranks are close is the same as the calculation in the case where two attribute information related to the same type are associated with the trajectory link candidate. Yes.
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood for each combination of attribute information having similar ranks related to the same type, then adds the attribute information likelihood, and adds the attribute information likelihood Perform normalization to calculate the power root of the number of degrees. In this way, the attribute information likelihood regarding the attribute information type of interest is calculated.
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood in the same manner for other attribute information types.
- the attribute information likelihood calculation unit 334 After calculating the attribute information likelihood for each attribute information type as described above, the attribute information likelihood calculation unit 334 adds the attribute information likelihoods, and calculates the root of the number of attribute information likelihoods used for the integration. Perform normalization. Thereby, the attribute information likelihood regarding the trajectory link candidate with attribute information in the case where attribute information of a plurality of attribute information types is associated with the trajectory link candidate is calculated. Further, when the attribute information likelihood calculated for each attribute information type is added, the attribute information likelihood may be multiplied by a weighting factor corresponding to the type, and may be added. By multiplying the attribute information likelihood by the weighting factor, the reliability of each attribute information type can be reflected in the attribute information likelihood.
- the gender of a person can be detected with high accuracy, but if there is an error in the detection of clothes color, the attribute information likelihood obtained by gender is multiplied by a large weighting factor to obtain the attribute information likelihood obtained for the color. Multiply by a small weighting factor.
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood for the first and third attribute information sets and the third and fourth attribute information sets. Thereby, since two attribute information likelihoods are obtained, the attribute information likelihood calculation unit 334 may add them and calculate the square root. Thereby, the attribute information likelihood regarding the attribute information type “color” is calculated.
- the attribute information likelihood regarding the attribute information type “gender” is calculated from the second and fifth attribute information. Thereafter, the attribute information likelihood calculation unit 334 adds the attribute information likelihood related to “color” and the attribute information likelihood related to “gender” to calculate a square root. Thereby, the attribute information likelihood regarding the locus
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood of the trajectory link candidate. It is assumed to be a constant (for example, “1”).
- the attribute information likelihood calculation unit 334 integrates and normalizes the attribute information likelihoods calculated for each trajectory link candidate belonging to the same hypothesis, and thereby calculates the attribute information likelihood of the hypothesis. In this normalization, a power root of the number of trajectory link candidates belonging to the same hypothesis is calculated.
- the attribute information likelihood calculation unit 334 integrates the trajectory link likelihood, the identification information likelihood, and the attribute information likelihood for each hypothesis. This integration process is performed by integrating the trajectory link likelihood, the identification information likelihood, and the attribute information likelihood. Further, when performing the integration process, the attribute information likelihood calculation unit 334 may weight and integrate the trajectory link likelihood, the identification information likelihood, and the attribute information likelihood.
- Trajectory link candidate generation unit 31, hypothesis generation unit 32d, and likelihood calculation unit 33d (likelihood calculation control unit 330d, trajectory link likelihood calculation unit 331, identification information likelihood calculation unit 332, and attribute information likelihood calculation unit 334 Is implemented by a CPU of a computer that operates in accordance with a moving body trajectory identification program, for example.
- the program storage device (not shown) of the computer stores the moving body trajectory identification program, and the CPU reads the program, and the functions of the trajectory link candidate generation unit 31, the hypothesis generation unit 32d, and the likelihood calculation unit 33d are performed. Realize.
- the trajectory link candidate generation unit 31, the hypothesis generation unit 32d, and the likelihood calculation unit 33d may be realized by separate hardware.
- the likelihood calculation control unit 330d, the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the attribute information likelihood calculation unit 334 may also be realized by separate hardware.
- FIGS. 30 and 31 are flowcharts illustrating the process of the moving object trajectory identification device 3d according to the fifth embodiment.
- the same steps as those in the first embodiment are denoted by the same reference numerals, and the description thereof is omitted.
- the processing until the table holding the acquired identification information is updated is the same as that of the first embodiment (see FIG. 4).
- the hypothesis generation unit 32d acquires a set of detection time, attribute information, attribute information type, position coordinate, and attribute information detection device number from the attribute information detection device 5 (step S51). When the attribute information detection device 5 detects the attribute information for the moving object, the attribute information detection device 5 outputs the attribute information to the hypothesis generation unit 32d.
- the hypothesis generation unit 32d holds a correspondence relationship between the detection time, the attribute information, the attribute information type, the position coordinates, and the attribute information detection device number in a table.
- the hypothesis generation unit 32d holds the attribute information acquired from the attribute information detection device 5 in the past fixed time from the current time (that is, the time when the latest attribute information is acquired).
- the hypothesis generation unit 32d acquires a set of the detection time, attribute information, attribute information type, position coordinate, and attribute information detection device number for the latest attribute information
- the hypothesis generation unit 32d additionally registers the set in the data, Old attribute information is deleted (step S52).
- the hypothesis generation unit 32d holds the correspondence relationship between the detection time, the attribute information, the attribute information type, the position coordinate, and the attribute information detection device number for the times t1 to tn-1.
- the hypothesis generation unit 32d stores the attribute information acquired at the time tn in the table. On the other hand, the attribute information acquired at time t1 is discarded from the table.
- the hypothesis generation unit 32d not only stores the updated table itself, but also outputs it to the likelihood calculation unit 33d.
- the hypothesis generation unit 32d uses the attribute information stored in the table updated in step S52 and the trajectory link candidate input from the trajectory link candidate generation unit 31 in step S3, that is, a combination of trajectory link candidate and attribute information (ie, attribute information). , A trajectory link candidate with attribute information) is generated (step S53). Specifically, the hypothesis generation unit 32d selects one piece of attribute information from the input attribute information, and calculates the Euclidean distance between the position coordinates of the locus connection candidate and the position coordinates of the selected attribute information at the detection time for all the loci. Calculation is performed for each combination of candidate for connection and selected attribute information.
- the hypothesis generation unit 32d specifies a combination of a trajectory link candidate and attribute information with the shortest Euclidean distance, and stores it as a trajectory link candidate with attribute information.
- the hypothesis generation unit 32d performs this process once for all the attribute information acquired in step S51.
- the hypothesis generation unit 32d interpolates the trajectory where the trajectory is interrupted, specifies the position coordinates of the trajectory link candidate, and calculates the Euclidean distance from the position coordinates of the attribute information. .
- the trajectory link candidate includes a plurality of trajectories at the detection time of the attribute information, and a plurality of position coordinates are detected, the Euclidean distance between the barycentric points and the position coordinates of the attribute information is calculated.
- one attribute information is not necessarily associated with one trajectory link candidate, and two or more attribute information may be associated with one trajectory link candidate. Alternatively, there may be a trajectory link candidate that is not associated with any attribute information.
- the hypothesis generation unit 32d uses the identification information stored in the table updated in step S5 and the trajectory link candidate with attribute information generated in step S53, and sets the trajectory link candidate with attribute information and the identification information ( Hereinafter, all “trajectory link candidate with attribute information / identification information pair” are generated (step S6d).
- the hypothesis generation unit 32d pays attention only to the trajectory link candidate in the trajectory link candidate with attribute information, and the trajectory link candidate / identification information with attribute information in the same manner as in step S6 (see FIG. 4) in the first embodiment. All pairs should be generated.
- the hypothesis generation unit 32d generates a hypothesis as a set of trajectory link candidate / identification information pairs with attribute information (step S7d).
- the hypothesis generation unit 32d pays attention to the trajectory link candidate and the identification information in the trajectory link candidate / identification information pair with attribute information, and the attributes satisfying the first condition to the third condition as in the first embodiment. All sets of trajectory link candidate / identification information pairs with information are extracted, and individual sets are assumed as hypotheses.
- the hypothesis generation unit 32d outputs the generated hypothesis group to the likelihood calculation unit 33d.
- the likelihood calculating unit 33d performs the processing after step S8 shown in FIG.
- the process of determining the presence or absence of a hypothesis whose likelihood has not been calculated and calculating the trajectory link likelihood and the identification information likelihood is the same as in the first embodiment.
- the likelihood calculation control unit 330d determines that there is an unexecuted hypothesis in the likelihood calculation process (specifically, step S9 to step S11d) in step S8 (that is, the determination result “ Yes "), one of the hypotheses whose likelihood is not calculated is output to the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the attribute information likelihood calculation unit 334.
- the calculation of the trajectory link likelihood and the identification information likelihood is the same as that in the first embodiment, and the description thereof is omitted.
- the attribute information likelihood calculation unit 334 calculates the attribute information likelihood for the input hypothesis (step S54).
- step S54 the attribute information likelihood calculation unit 334 calculates the attribute information likelihood for each trajectory link candidate included in the hypothesis, normalizes the attribute information likelihood for each trajectory link candidate belonging to the same hypothesis, Thus, the attribute information likelihood is calculated.
- the attribute information used for calculating the attribute information likelihood is provided from the hypothesis generation unit 32 as a set of the updated detection time (that is, after step S52), the attribute information, and the attribute information type.
- the attribute information likelihood calculation unit 334 calculates the hypothesis trajectory link likelihood calculated in step S9, the hypothesis identification information likelihood calculated in step S10, and the calculation in step S54.
- the attribute information likelihoods are integrated, and the integration results and hypotheses are stored in association with each other (step S11d).
- step S11d integration is performed by integrating the trajectory link likelihood, identification information likelihood, and attribute information likelihood calculated in steps S9, S10, and S54.
- the attribute information likelihood calculation unit 334 may weight the trajectory link likelihood, the identification information likelihood, and the attribute information likelihood.
- the likelihood calculating unit 33d repeats the loop from step S8 to step S11d.
- the attribute information calculation unit 334 causes the trajectory link likelihood, the identification information likelihood, and the attribute information likelihood.
- the hypothesis with the maximum integrated value is estimated as the maximum likelihood hypothesis (step S12d).
- the attribute information likelihood calculation unit 334 sends the estimated maximum likelihood hypothesis to the identification result output device 4.
- the identification result output device 4 When the identification result output device 4 receives the maximum likelihood hypothesis estimated in step S12d, the identification result output device 4 refers to the trajectory link candidate / identification information pair with attribute information belonging to the maximum likelihood hypothesis, and which trajectory is which mobile body. It is determined whether it is a locus (step S13). At this time, the identification result output device 4 may specifically display the position or locus of the moving body. Step S13 is the same as that in the first embodiment.
- the attribute information detecting device 5 detects the feature of the moving object as attribute information
- the moving object trajectory identifying device 3d uses the attribute information likelihood as the likelihood of the hypothesis in addition to the trajectory link likelihood and the identification information likelihood. Use for calculation.
- the calculation results of the trajectory link likelihood and the identification information likelihood in the plural trajectory link candidate / identification information pairs are substantially the same, and there is a difference in the temporary likelihood. It may be assumed that it does not occur.
- attribute information likelihood is also used in the likelihood calculation of the hypothesis in the fifth embodiment, even if a plurality of moving bodies move along similar routes, if there are differences in the characteristics of the moving bodies, the hypothesis Since there is a difference in likelihood, it is possible to estimate which trajectory is the trajectory of which moving body. That is, attribute information likelihood is also integrated in the likelihood integration process (step S11d), and the maximum likelihood hypothesis is estimated according to the integration result. Can do.
- the likelihood calculation unit 33d in the fifth embodiment may not include the trajectory link likelihood calculation unit 331 and the trajectory link likelihood may not be calculated.
- the attribute information likelihood calculation unit 334 integrates only the identification information likelihood and the attribute information likelihood, and estimates the hypothesis having the highest integrated value as the maximum likelihood hypothesis.
- the likelihood calculator 33d does not include the trajectory link likelihood calculator 331 in the fifth embodiment, and the mobile trajectory identification device 3d includes the trajectory link likelihood calculator 331a. May be.
- the attribute information likelihood calculation unit 334 integrates only the identification information likelihood and the attribute information likelihood, and estimates the hypothesis having the highest integrated value as the maximum likelihood hypothesis. With this configuration, the same effects as in the second embodiment can be obtained in the fifth embodiment.
- Example 5 or its modification example and Example 3 may be combined. Or you may combine Example 5 or its modification, and Example 4.
- FIG. Further, the fifth embodiment or its modification may be combined with the fourth and third embodiments.
- FIG. 32 is a block diagram showing a moving body trajectory identification system according to Embodiment 6 of the present invention.
- the moving body locus identification system according to the sixth embodiment includes a position information detection device 1, an identification information detection device 2, a moving body locus identification device 3e, an identification result output device 4, and a movement information detection device 6.
- the moving body trajectory identification device 3e includes a trajectory link likelihood generation unit 31, a hypothesis generation unit 32e, a likelihood calculation unit 33e, and a map storage unit 37.
- the number of movement information detection devices 6 included in the moving body locus identification system is not limited to one, and a plurality of movement information detection devices 6 may be included.
- a movement information detection device ID is assigned in advance. Note that the movement information detection device ID is represented by a number and is referred to as a movement information detection device number.
- the movement information detection device 6 acquires movement information of the moving body in the tracking area 50.
- the movement information is information indicating the movement status of the mobile object that is the tracking target, and means, for example, the movement speed, the movement direction, and the like.
- the movement information detection device 6 may be any device that can detect movement information of a moving body in the tracking area 50 and specify the detection time, and is realized by, for example, a camera, a laser range finder, an acceleration sensor, or the like.
- the movement information detection device 6 may be installed in the tracking area 50. Or you may make it a mobile body itself carry the portable terminal carrying the sensor which can detect movement information.
- the detection areas of the movement information detection devices 6 may be installed so as to overlap each other. Or you may install so that the detection area
- the movement information detection device 6 is not necessarily capable of detecting the movement information, and whether the movement information is detected depends on the installation state of the movement information detection device 6. For example, when the movement information detection device 6 is installed in the tracking area, the movement information is easy to be detected if the movement object exists near the movement information detection device 6, but the movement object is detected by the movement information detection device 6. If it is present at a location away from the location, the movement information is difficult to detect. Further, when the mobile body itself carries a mobile terminal equipped with the movement information detection device 6 such as an acceleration sensor, that is, when the movement information detection device 6 is attached to the mobile body, the movement information detection device 6 moves. Body movement information can always be detected.
- the position coordinates where the movement information is detected are not necessarily detected.
- the movement information detection device 6 is an acceleration sensor mounted on a portable terminal, and detects a moving state of a moving body such as “still”, “walking”, “running”, or “ When detecting the “number of steps”, the movement information detection device 6 does not detect the position coordinates where the movement information is detected. In this case, the movement information detection device 6 determines that the position coordinate where the movement information is detected is “none”. In the following description, when the position coordinates of the movement information are not detected, the position coordinates are represented by a character string “none”. Note that this position coordinate may be represented by another character string.
- the movement information detection device 6 detects the movement information of the moving body and also detects the position coordinates of the moving body when the movement information is detected. Movement information in which specific values are detected as position coordinates is referred to as “movement information with position coordinates”.
- the movement information detection device 6 uses the detected movement information, the movement information type uniquely representing the type, the detection time of the movement information, the position coordinates of the movement information, and the movement information detection device number as the moving body trajectory identification information. Send to 3e.
- the position information of the movement information is “none”. If the movement information detection device 6 tries to detect the movement information but fails to detect it, the movement information and the movement information type are set to “none”, and the moving object trajectory identification information 3e together with the time and the movement information detection device number. May be sent to Alternatively, the movement information detection device 6 may determine that no movement information has been detected at that time because the movement information detection device 6 does not send the movement information to the movement object locus identification information 3e.
- the position coordinates of the movement information detection device 6 are moved. It is sent to the body locus identification information 3e.
- the movement information detection device 6 may hold the position coordinates of its own installation position in the tracking area 50 in advance.
- the movement information detection device 6 detects movement information from an image captured using a camera
- the range of the captured image is wide, so the movement information is extracted from the installation position of the movement information detection device 6 and the image. It does not necessarily match the position coordinates of the moving object at the time.
- the movement information detection device 6 stores in advance a conversion table for converting an arbitrary position in the captured image into a position coordinate in the tracking area 50, and the position coordinate of the moving body in the image is stored in the tracking area 50 by the conversion table. It may be converted into the position coordinates within.
- the position coordinates of the moving body at the time of detecting the movement information may be obtained by a method according to the type of movement information detection device 6 (in other words, the movement information type).
- the position information detection device 1 detects the position information of the moving body
- the identification information detection device 2 detects the identification information of the moving body
- the movement information detection device 6 detects the movement information.
- the position information detection device 1, the identification information detection device 2, and the movement information detection device 6 detect the position information, identification information, and movement information asynchronously
- the moving object locus identification device 3e has the position information, the identification device, and
- the movement information may be buffered for a predetermined time, and the position information, the identification device, and the movement information stored in the buffer may be used every fixed time.
- the moving object locus identification device 3e receives the position information, the identification device, and the movement that are simultaneously input.
- the same detection time may be set for the information.
- the hypothesis generation unit 32e inputs a trajectory link candidate from the trajectory link candidate generation unit 31 and inputs identification information from the identification device detection device 2.
- the hypothesis generation unit 32e inputs movement information, movement information type, detection time, position coordinates, and movement information detection apparatus number from the movement information detection apparatus 6, and holds the corresponding relationship.
- the hypothesis generation unit 32e generates all hypotheses based on the trajectory link candidate input from the trajectory link candidate generation unit 31, the identification information input from the identification information detection device 2, and the movement information input from the movement information detection device 6. Then, the generated hypothesis group is output to the likelihood calculating unit 33e.
- the hypothesis generation unit 32e uses the trajectory link candidate input from the trajectory link candidate generation unit 31 and the movement information detected by the movement information detection device 6 in the time zone from the current time to a time that is a certain time in the past. A pair of trajectory link candidates and movement information (that is, trajectory link candidates with movement information) is generated.
- the hypothesis generation unit 32e selects a trajectory link candidate that associates movement information according to the following procedure, and generates a trajectory link candidate with movement information.
- the hypothesis generation unit 32e associates movement information having position coordinates among movement information input from the movement information detection device 6 with a trajectory link candidate.
- the hypothesis generation unit 32e selects only the movement information having the position coordinates from the movement information input from the movement information detection device 6, and selects one movement information from the movement information.
- the hypothesis generation unit 32e calculates the Euclidean distance between the position coordinates of the trajectory link candidate at the detection time of the selected movement information and the position coordinates of the selected movement information for all the trajectory link candidates and the selected set of movement information.
- the hypothesis generation unit 32e specifies and stores a combination of the trajectory link candidate and the movement information with the shortest Euclidean distance. This process is performed once for all pieces of movement information input from the movement information detection device 6.
- the trajectory link candidate associated with the movement information by this process is referred to as “provisional trajectory link candidate with movement information”.
- the hypothesis generation unit 32e interpolates the interruption locus and connects the locus at the detection time of the movement information. It is only necessary to identify the position coordinates of the candidate and calculate the Euclidean distance from the position coordinates of the movement information. If the trajectory link candidate includes a plurality of trajectories at the detection time of the movement information and there are a plurality of position coordinates of the moving object at the detection time, the Euclidean distance between the center of gravity and the position coordinates of the movement information is set. Calculate it. This process is the same as the process of generating the trajectory link candidate with attribute information in the fifth embodiment.
- the hypothesis generation unit 32e associates the remaining movement information with the position coordinate “none” among the movement information input from the movement information detection device 6 and the trajectory link candidate with temporary movement information.
- the hypothesis generation unit 32e selects only the movement information having the position coordinate “none” from the movement information input from the movement information detection apparatus 6, and connects the selected movement information and the trajectory with temporary movement information. All pairs with candidates are generated. That is, a combination of the movement information of the position coordinate “none” and the trajectory link candidate with temporary movement information is the official “trajectory link candidate with movement information”. This process means adding movement information of position coordinates “none” as movement information associated with the trajectory link candidate.
- the temporary trajectory link candidate with movement information is regarded as an official “trajectory link candidate with movement information”. Determined.
- trajectory link candidate associated with one piece of movement information, or there may be a trajectory link candidate associated with a plurality of pieces of movement information.
- trajectory link candidate that is not associated with movement information.
- trajectory link candidates T1, T2,..., Tn are generated, movement information M1, M2,..., Mm having position coordinates are detected, and movement information with position coordinates “none” does not exist.
- the fact that there is no movement information associated with the trajectory link candidate is expressed by a character string “no movement information”.
- the hypothesis generation unit 32e includes (T1, M1), (T2, M2),..., (Tn, Mm), (T1, no movement information), (T2, M1), (T3, M2, M3),..., (Tn, Mm-2, Mm-1, Mm) are generated (that is, a trajectory link candidate with temporary movement information).
- the trajectory link candidate with temporary movement information is held as a trajectory link candidate with formal movement information.
- the hypothesis generation unit 32e may generate a trajectory link candidate with movement information by associating all input movement information with any trajectory link candidate. Further, when the hypothesis generation unit 32e generates a temporary trajectory link candidate with movement information, the Euclidean distance is determined in advance for the trajectory link candidate that is determined to have the shortest Euclidean distance from the position coordinates of the movement information. If the threshold value is exceeded, it may be determined that the movement information is not associated with any trajectory link candidate.
- the hypothesis generation unit 32e uses the trajectory link candidate with movement information generated according to the above procedure, and the identification information detected by the identification information detection device 2 in the time zone from the current time to a time that is a certain time past in the past, All sets of trajectory link candidates with movement information and identification information are generated.
- the hypothesis generation unit 32e focuses on only the trajectory link candidate in the trajectory link candidate with movement information and identifies the trajectory link candidate with movement information in the same process as the generation process of the trajectory link candidate / identification information pair in the first embodiment. A complete set of information may be generated.
- a combination of a trajectory link candidate with movement information and identification information is referred to as a trajectory link candidate with movement information / identification information pair.
- the hypothesis generation unit 32e extracts all sets of trajectory link candidate / identification information pairs with movement information that satisfy the first to third conditions, and sets each set as a hypothesis.
- the likelihood calculation unit 33e calculates and integrates the trajectory link likelihood, the identification information likelihood, and the movement information likelihood for each hypothesis generated by the hypothesis generation unit 32e.
- the hypothesis having the largest integrated value is estimated as the maximum likelihood hypothesis and sent to the identification result output device 4.
- the movement information likelihood is a moving object in which the movement information associated with the locus connection candidate and the movement state of the moving object derived from the locus connection candidate in the locus connection candidate with movement information included in the hypothesis are the same. It represents the likelihood that it represents.
- the moving information likelihood may be represented by the similarity between the moving body speed obtained from the moving information and the moving body speed obtained from the trajectory link candidate.
- the movement information likelihood may be represented by the similarity between the moving direction of the moving object obtained from the moving information and the moving direction of the moving object obtained from the trajectory link candidate.
- FIG. 33 is a block diagram illustrating a configuration of a likelihood calculating unit 33e according to the sixth embodiment.
- the likelihood calculation unit 33e includes a likelihood calculation control unit 330e, a trajectory link likelihood calculation unit 331, an identification information likelihood calculation unit 332, and a movement information likelihood calculation unit 335.
- the trajectory link likelihood calculation unit 331 and the identification information likelihood calculation unit 332 are the same as those in the first embodiment, description thereof is omitted.
- the likelihood calculation control unit 330e outputs a hypothesis whose likelihood has not been calculated to the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the movement information likelihood calculation unit 335.
- the movement information likelihood calculation unit 335 receives the hypothesis generated by the hypothesis generation unit 32e and calculates the movement information likelihood for each hypothesis. When calculating the individual movement information likelihood, the movement information likelihood calculating unit 335 calculates the movement information likelihood for each trajectory link candidate included in the hypothesis, and calculated for each trajectory link candidate belonging to the same hypothesis.
- the movement information likelihood is integrated and normalized to calculate the movement information likelihood of the hypothesis.
- the movement information likelihood calculation unit 335 holds in advance a likelihood calculation method for each movement information type, and determines the calculation method of the movement information likelihood based on the type of movement information included in the hypothesis. Then, the movement information likelihood calculation unit 335 calculates the movement information likelihood by a calculation method according to the movement information type.
- one type of movement information type may be associated with the trajectory link candidate, or a plurality of types of movement information types may be associated.
- the movement information likelihood calculation unit 335 refers to the movement information associated with the trajectory link candidate.
- the movement information indicates a vector V1 representing the moving direction of the moving body.
- the movement information likelihood calculation unit 335 calculates a vector V2 representing the moving direction of the moving object at the movement information detection time based on the trajectory link candidate.
- the movement information likelihood calculation unit 335 may calculate the vector V2 based on, for example, the position coordinates of the moving body before and after the movement information detection time.
- the movement information likelihood calculation unit 335 calculates the cosine similarity of the two vectors V1 and V2 to obtain the movement information likelihood.
- the movement information likelihood calculation unit 335 calculates the cosine similarity “cos ⁇ ” by performing calculation according to Equation 1 using the two vectors V1 and V2.
- the cosine similarity is calculated and the movement direction vector similarity is calculated to obtain the movement information likelihood.
- the movement direction vector similarity may be calculated by another calculation method.
- the movement information likelihood calculation unit 335 refers to the speed associated with the trajectory link candidate as the movement information.
- the movement information representing the speed may be a numerical value representing the movement speed.
- the movement information indicating the speed may be a character string such as “stationary” indicating that the vehicle is stationary, “walking” indicating the walking speed, and “running” indicating the speed faster than walking.
- the movement information represented by a numerical value or a character string is output from the movement information detection apparatus 6.
- the movement information likelihood calculation unit 335 extracts the time before and after the detection time of the movement information and the position coordinates of the moving object at that time from the trajectory link candidate, and calculates the speed of the moving object.
- the movement information likelihood calculation unit 335 calculates the difference between the speed obtained from the movement information and the speed calculated from the trajectory link candidate.
- a likelihood function is defined in advance such that the moving information likelihood increases as the speed difference decreases, and the moving information likelihood decreases as the speed difference increases, and is stored in the moving information likelihood calculating unit 335.
- the movement information likelihood calculation unit 335 calculates the movement information likelihood by substituting the difference between the speed obtained from the movement information and the speed calculated from the trajectory link candidate into the likelihood function.
- the movement information likelihood calculating unit 335 determines the movement information likelihood in the following procedure.
- the movement information likelihood calculation unit 335 classifies the speed calculated from the trajectory link candidate into one of the speed categories.
- the speed range in each speed section is preset.
- the movement information likelihood calculation unit 335 determines the movement information likelihood based on the combination of the speed classification indicated by the movement information and the speed classification determined based on the speed calculated from the trajectory link candidate. In this case, the movement information likelihood is defined in advance for each combination of speed categories, and is stored in the movement information likelihood calculation unit 335.
- speed classification of movement information such as [stationary, stationary], [stationary, walking], [stationary, running], [walking, walking], [walking, running], [running, running], and trajectory connection candidates
- the movement information likelihood is defined in advance for all the combinations of speed categories calculated from the above, and is stored in the movement information likelihood calculation unit 335.
- the moving information likelihood is set to be higher as the contents of the speed section are approximated, and the moving information likelihood is set to be lower as the contents of the speed section are different.
- the movement information likelihood is set high, and the speed is greatly different as in [stationary, traveling].
- the movement information likelihood is set low. Moreover, it is only necessary to define a medium movement information likelihood for a combination that does not have the same speed category as [Still, Walk] or [Walk, Run] but does not differ as much as [Still, Run]. . Then, the movement information likelihood calculation unit 335 selects the movement information likelihood corresponding to the combination of the speed classification indicated by the movement information and the speed classification calculated from the trajectory link candidate from the predefined movement information likelihoods.
- the above example is a case where one piece of movement information is associated with the trajectory link candidate.
- the movement information likelihood calculation unit 335 calculates the movement information likelihood related to the trajectory link candidate according to the following procedure. First, the case where two pieces of movement information of the same type are associated with the trajectory link candidate will be described.
- the movement information likelihood calculation unit 335 orders the movement information associated with the trajectory link candidate in the trajectory link candidate with movement information in order of early detection time.
- the movement information likelihood calculation unit 335 calculates the movement information likelihood based on the movement information and the trajectory link candidate in the order of detection time.
- the calculation process of the movement information likelihood based on the individual movement information and the trajectory link candidate is the same as the calculation process in the case where one movement information is associated with the trajectory link candidate.
- the movement information likelihood calculation unit 335 calculates the movement information likelihood using the trajectory link candidate for all movement information, and then accumulates the movement information likelihood and relates to the number of accumulated movement information likelihoods. Normalization for obtaining a power root is performed, and thus the movement information likelihood of the trajectory link candidate with movement information is calculated.
- each piece of movement information is weighted and integrated in accordance with the movement information detecting device number corresponding to the original movement information for which individual movement information likelihood is calculated. May be. That is, a large weight is given to the movement information likelihood calculated based on the movement information input from the movement information detection apparatus with high detection accuracy, while, based on the movement information input from the movement information detection apparatus with low detection accuracy.
- the movement information likelihood may be integrated after giving a small weight to the calculated movement information likelihood. If the weighting according to the movement information detection device number is not performed, the movement information detection device 6 may not output the movement information detection device number.
- the movement information likelihood calculation unit 335 calculates the movement information likelihood for each movement information type.
- the process of calculating the movement information likelihood for each movement information type is the same as the calculation process for the trajectory link candidate associated with the same type of movement information.
- the movement information likelihood calculation unit 335 integrates the movement information likelihood and performs normalization to obtain a power root of the number of the accumulated movement information likelihoods.
- the movement information likelihood can be calculated for the trajectory link candidate with movement information that includes the trajectory link candidates associated with two or more types of movement information.
- the movement information likelihood may be integrated by multiplying each movement information likelihood by a weighting factor corresponding to the movement information type. .
- the movement information likelihood calculation unit 335 sets the movement information likelihood of the trajectory link candidate as a constant (for example, “1”).
- the movement information likelihood calculation unit 335 sets the movement information likelihood calculated for each trajectory link candidate belonging to the same hypothesis as a movement information likelihood of the hypothesis. In this normalization, a power root of the number of trajectory link candidates belonging to the same hypothesis is calculated.
- the movement information likelihood calculation unit 335 integrates the trajectory link likelihood, the identification information likelihood, and the movement information likelihood for each hypothesis. This integration process is performed by integrating the trajectory link likelihood, the identification information likelihood, and the movement information likelihood. In this integration process, the movement information likelihood calculation unit 335 may perform weighting and integration on the trajectory link likelihood, the identification information likelihood, and the movement information likelihood.
- the trajectory link candidate generation unit 31, the hypothesis generation unit 32e, the likelihood calculation unit 33e (the likelihood calculation control unit 330e, the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the movement information likelihood calculation unit 335 For example) is realized by a CPU of a computer that operates in accordance with the moving object trajectory identification program.
- the program storage device (not shown) of the computer stores the moving body trajectory identification program, and the CPU reads the program, and the trajectory link candidate generation unit 31, the hypothesis generation unit 32e, and the likelihood calculation unit 33e according to the program. Realize the function.
- the trajectory link candidate generation unit 31, the hypothesis generation unit 32e, and the likelihood calculation unit 33e may be calibrated with separate hardware.
- the likelihood calculation control unit 330e, the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the movement information likelihood calculation unit 335 may be configured by separate hardware.
- FIG. 34 and FIG. 35 are flowcharts illustrating processing of the moving object trajectory identification device 3e according to the sixth embodiment.
- the same steps as those in the first embodiment are denoted by the same reference numerals, and the description thereof is omitted.
- the processing until the table holding the identification device in FIG. 34 is updated is the same as that of the first embodiment.
- the hypothesis generation unit 32e acquires a set of detection time, movement information, movement information type, position coordinates, and movement information detection device number from the movement information detection device 6 (step S61).
- the movement information detection device 6 detects movement information of the moving body and outputs the movement information to the hypothesis generation unit 32e.
- the hypothesis generation unit 32e holds a correspondence relationship between the detection time, the movement information, the movement information type, the position coordinates, and the movement information detection device number in a table.
- the hypothesis generation unit 32e holds the movement information acquired from the movement information detection device 6 in the past fixed time from the current time (that is, the latest detection time of the input movement information).
- the hypothesis generation unit 32e acquires a set of the latest detection time, movement information, movement information type, position coordinates, and movement information detection device number, additionally registers it in the table, and deletes the oldest movement information (step S62). ).
- the hypothesis generation unit 32e holds the correspondence relationship between the detection time, the movement information, the movement information type, the position coordinates, and the movement information detection device number for the times t1 to tn-1.
- the hypothesis generation unit 32e stores the movement information acquired for the latest time tn in a table.
- the movement information at the oldest time t1 is discarded from the table.
- the hypothesis generation unit 32e not only stores the updated table itself, but also outputs it to the likelihood calculation unit 33e.
- the hypothesis generation unit 32e uses the movement information stored in the table updated in step S62 and the trajectory link candidate input from the trajectory link candidate generation unit 31 in step S3, thereby using the trajectory link candidate with temporary movement information. (That is, a set of trajectory link candidates and movement information having position coordinates) is generated (step S63). Specifically, the hypothesis generation unit 32e selects only movement information having position coordinates from the input movement information, and selects one movement information having position coordinates from the selected movement information. Then, the hypothesis generation unit 32e calculates the Euclidean distance between the trajectory link candidate and the position coordinates of the selected movement information at the detection time for all the trajectory link candidates and the selected movement information set.
- the hypothesis generation unit 32e specifies a combination of the trajectory link candidate and the movement information with the shortest Euclidean distance, and sets it as a temporary trajectory link candidate with movement information. This process is performed once for all movement information having position coordinates. As described above, in the case of a trajectory link candidate where the trajectory is interrupted, the hypothesis generation unit 32e interpolates the trajectory where the trajectory is interrupted to identify the position coordinates for the trajectory link candidate, and thus the position coordinates of the movement information Calculate the Euclidean distance of. Further, when a plurality of trajectories detected at the movement information detection time are included in the trajectory link candidate and there are a plurality of position coordinates, the Euclidean distance between the barycentric point and the position coordinates of the movement information is calculated.
- step S63 only one piece of movement information having position coordinates is associated with one locus connection candidate, and two or more pieces of movement information having position coordinates are associated with one locus connection candidate. It may be. Further, there may be a trajectory link candidate that is not associated with movement information having position coordinates.
- the hypothesis generation unit 32e associates the movement information with the position coordinate “none” (that is, movement information not selected in step S63) and the trajectory link candidate with temporary movement information (step S64).
- the hypothesis generation unit 32e generates a complete set of the movement information having the position coordinate “none” and the individual trajectory link candidates with temporary movement information, and forms the formal “trajectory link with movement information”.
- Candidate ". If there is no movement information with the position coordinate “none”, the hypothesis generation unit 32e sets a temporary trajectory link candidate with movement information as a formal trajectory link candidate with movement information.
- the hypothesis generation unit 32e uses the identification information stored in the table updated in step S5 and the trajectory link candidate with movement information generated in step S64 to complete the trajectory link candidate / identification information pairs with movement information. Generate (step S6e).
- the hypothesis generation unit 32e pays attention only to the trajectory link candidates in the trajectory link candidates with movement information, and may generate all the trajectory link candidates with movement information / identification information pairs as in step S6 of the first embodiment.
- the hypothesis generation unit 32e generates a hypothesis as a set of trajectory link candidate / identification information pairs with movement information (step S7e).
- the hypothesis generation unit 32e pays attention to the trajectory link candidate and the identification information in the trajectory link candidate / identification information pair with movement information, and the movement satisfying the first condition to the third condition as in the first embodiment. All sets of trajectory link candidate / identification information pairs with information are extracted, and individual sets are assumed as hypotheses.
- the hypothesis generation unit 32e outputs the generated hypothesis group to the likelihood calculation unit 33e.
- the likelihood calculating unit 33e performs the processing after step S8 shown in FIG.
- the processing i.e., steps S8 to S10) from determining the presence or absence of a hypothesis whose likelihood has not been calculated until calculating the trajectory link likelihood and the identification information likelihood is the same as in the first embodiment.
- the likelihood calculation control unit 330e determines in step S8 that there is a hypothesis for which the likelihood is not calculated (that is, a hypothesis in which the processes in steps S9 to S11e are not executed) (the determination result “Yes” in step S8). )
- One of the hypotheses whose likelihood has not been calculated is output to the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the movement information likelihood calculation unit 335.
- the calculation of the trajectory link likelihood and the identification information likelihood is the same as that in the first embodiment, the description thereof is omitted.
- the movement information likelihood calculation unit 335 calculates the movement information likelihood for the input hypothesis (step S65). In step S65, the movement information likelihood calculation unit 335 calculates the movement information likelihood for each trajectory link candidate included in the hypothesis, adds and normalizes the movement information likelihood for each trajectory link candidate belonging to the same hypothesis, Thus, the movement information likelihood of the hypothesis is calculated.
- the movement information used for calculating the movement information likelihood is provided from the hypothesis generation unit 33e as a set of the detection time after the update (that is, after step S62), the movement information, and the movement information type.
- the movement information likelihood calculation unit 335 calculates the hypothesis trajectory link likelihood calculated in step S9, the hypothesis identification information likelihood calculated in step S10, and the calculation in step S65.
- the movement information likelihoods are integrated, and the integration results and hypotheses are stored in association with each other (step S11e).
- the trajectory link likelihood, the identification information likelihood, and the movement information likelihood calculated in steps S9, S10, and S65 are integrated and integrated.
- the movement information likelihood calculation unit 335 may weight the trajectory link likelihood, the identification information likelihood, and the movement information likelihood.
- the likelihood calculating unit 33e repeats the loop from step S8 to step S11e.
- the movement information likelihood calculation unit 335 determines the trajectory link likelihood, the identification information likelihood, and the movement information likelihood.
- the hypothesis with the maximum integration result is estimated as the maximum likelihood hypothesis (step S12e).
- the movement information likelihood calculation unit 335 sends the estimated maximum likelihood hypothesis to the identification result output device 4.
- Step S13 When the identification result output device 4 inputs the maximum likelihood hypothesis estimated in step S12e, based on the trajectory link candidate / identification information pair with movement information belonging to the maximum likelihood hypothesis, which trajectory is the trajectory of which mobile body. Is determined (step S13).
- the identification result output device 4 may specifically display the position coordinates or locus of the moving body. Step S13 is the same as that in the first embodiment.
- the moving information detecting device 6 detects the moving information of the moving object, and the moving object trajectory identifying device 3e calculates the moving object trajectory likelihood in addition to the trajectory link likelihood and the identification information likelihood, and the likelihood calculation of the hypothesis.
- the trajectory link likelihood and the identification information likelihood alone do not give a superiority or inferiority to the hypothesis likelihood, and it is not possible to accurately estimate which trajectory is which trajectory, if there is a difference in the movement status of the mobile
- the movement information a difference occurs in the likelihood of the hypothesis, and thus it is possible to accurately estimate which trajectory is which trajectory. That is, since the movement information likelihood is integrated in step S11e and the maximum likelihood hypothesis is estimated using the integration result, it is possible to estimate with high accuracy which locus is the locus of the moving body.
- the modification of the first embodiment may be applied to the sixth embodiment, and the likelihood calculating unit 33d may not include the trajectory link likelihood calculating unit 331 and may not calculate the trajectory link likelihood.
- the movement information likelihood calculation unit 335 may integrate only the identification information likelihood and the movement information likelihood and estimate the hypothesis having the highest integrated value as the maximum likelihood hypothesis.
- the likelihood calculation unit 33e does not include the trajectory link likelihood calculation unit 331, and the mobile trajectory identification device 3e includes the trajectory link likelihood calculation unit 331a as in the second embodiment. It is good.
- the trajectory link likelihood calculation unit 335 integrates only the identification information likelihood and the movement information likelihood, and estimates the hypothesis having the highest integrated value as the maximum likelihood hypothesis. With this configuration, the same effect as in the second embodiment can be obtained.
- the likelihood calculation unit 33e is combined with the trajectory link likelihood calculation unit 331, the identification information likelihood calculation unit 332, and the movement information likelihood calculation unit 335 in combination with the sixth embodiment and the fifth embodiment. You may make it comprise the calculation part 334 (refer FIG. 29).
- the moving body trajectory identification system also includes an attribute information detection device 5.
- the integration process of the likelihood, the trajectory link likelihood, the identification information likelihood, the movement information likelihood, and the attribute information likelihood are integrated, and the maximum likelihood hypothesis is estimated based on the integration result.
- FIG. 36 is a block diagram showing a moving object trajectory identification system according to Example 7 of the present invention.
- the moving body locus identification system according to the seventh embodiment includes a position information detection device 1, an identification information detection device 2, a moving body locus identification device 3f, and an identification result output device 4.
- the moving body trajectory identification device 3f according to the seventh embodiment includes a trajectory link candidate generation unit 31, a hypothesis generation unit 32f, a likelihood calculation unit 33, a map storage unit 37, and an environment information storage unit 38.
- the environment information storage unit 38 stores, as environment information, the location of the obstacle present in the tracking area 50 and an indicator of the travel time when the moving object passes through the location where the obstacle exists.
- the location of the obstacle and the indicator of the travel time when the mobile body passes through the location are the environmental information, but other elements may be the environmental information. Other examples of environmental information will be described later.
- an index of travel time when a moving object passes through a place where an obstacle exists is referred to as “cost”.
- the hypothesis generation unit 32 f receives the trajectory link candidate from the trajectory link candidate generation unit 31 and the identification information from the identification information detection device 2. In addition to this, the hypothesis generation unit 32 f reads environment information from the environment information storage unit 38.
- the hypothesis generation unit 32f calculates the moving time of the moving object while the trajectory is disconnected in the trajectory connection candidate based on the disappearance time of the trajectory and the generation time of the trajectory for the trajectory connection candidate in which the trajectory is interrupted. To do.
- this travel time is referred to as an actual travel time.
- the hypothesis generation unit 32f calculates the moving distance of the moving object in the actual moving time, and estimates the moving time for the moving object based on the environment information.
- this estimated value is referred to as an estimated travel time. If the estimated travel time is longer than the actual travel time, the trajectory link candidate is rejected from the trajectory link candidate group used for hypothesis generation. That is, the hypothesis generation unit 32f generates a hypothesis using a trajectory link candidate whose estimated travel time is shorter than the actual travel time, and a trajectory link candidate in which a trajectory is not interrupted.
- FIG. 37 is a diagram schematically showing environment information.
- the tracking area 50 is divided into a plurality of areas, and the cost determined for each divided area is stored in the environment information storage unit 38 as environment information.
- each divided area is referred to as a cell.
- the cost of the area where no obstacle exists (cells shown in white in FIG. 37) is set to “0”.
- a cell that takes a long time for the moving object to pass due to the presence of an obstacle sets its cost to a higher value.
- FIG. 37 the cost of the area where no obstacle exists
- the cost is set to “3.0” in the cell in the central area 61 where the obstacle exists, and the cost is set to “0.5” in the cell in the outer area 62 of the area where the obstacle exists. It also means that there is an obstacle in a cell whose cost is not “0”.
- the environmental information set in this way is referred to as an obstacle map.
- the hypothesis generation unit 32f calculates the time from the disappearance time of the trajectory in the trajectory link candidate to the occurrence time of the next trajectory as the actual travel time.
- the hypothesis generation unit 32f interpolates a straight line between the position coordinates of the locus disappearance in the locus connection candidate and the position coordinates of the next locus generation.
- the hypothesis generation unit 32f is the position coordinates on the interpolated straight line, and at the detection time (that is, the detection time of the position information and the identification information) in the actual movement time (in other words, the time when the trajectory is interrupted). Calculate position coordinates.
- the hypothesis generation unit 32f calculates the sum of the costs of the cells corresponding to each position coordinate.
- the hypothesis generation unit 32f calculates the distance from the disappearance position of the trajectory in the trajectory link candidate to the generation position of the next trajectory, and regards this distance as the moving distance of the moving object while the trajectory is interrupted.
- the hypothesis generation unit 32f divides the moving distance of the moving object while the trajectory is interrupted by the upper limit value of the moving speed of the moving object, adds the total cost to the division result, and calculates the addition result as the estimated moving time.
- the upper limit value of the moving speed of the moving body is determined in advance and is stored in the temporary generation unit 32f.
- the hypothesis generation unit 32f performs the above-described process on the trajectory link candidate where the trajectory is interrupted. Then, trajectory link candidates whose estimated travel time is longer than the actual travel time are excluded from trajectory link candidates used for hypothesis generation.
- trajectory link candidates including trajectories a and b shown in FIG. It is assumed that the disappearance time of the locus a is t1, and the occurrence time of the locus b is t6. Here, time t1 is earlier than time t6. Further, it is assumed that the disappearance position coordinates of the trajectory a are (Xa, Ya) and the generation position coordinates of the trajectory b are (Xb, Yb).
- the hypothesis generation unit 32f calculates an actual movement time t6-t1 in which the locus is interrupted.
- the hypothesis generation unit 32f interpolates a straight line between the disappearance position coordinate (Xa, Xb) of the trajectory a and the generation position coordinate (Xb, Yb) of the trajectory b, and is the position coordinate on the straight line.
- the position coordinates at the detection times t2 to t5 within the actual movement time are calculated.
- the costs of the cells corresponding to the positions at times t2, t3, t4, and t5 are “0”, “0.5”, “3.0”, and “0.5”.
- the hypothesis generation unit 32f calculates the moving distance L of the moving body while the locus is interrupted using Equation 5.
- the hypothesis generation unit 32f calculates the estimated movement time Tm using Equation 6.
- Equation 6 L is the movement distance calculated in Equation 5, and Vmax is a value set in advance as the upper limit value of the moving speed of the moving body.
- C is the total cost in the interpolated space, and is “4.0” in FIG.
- the hypothesis generation unit 32f rejects the trajectory link candidate including the trajectories a and b if the estimated travel time Tm calculated by Expression 6 is larger than the actual travel time t6-t1.
- the cost in FIG. 38 has a role of correcting so as to lengthen the travel time when passing through the path when there is an obstacle in the path interpolated between the trajectories a and b.
- the hypothesis generation unit 32f For the trajectory link candidates left without being rejected, a straight line is interpolated between the interrupted trajectories, and the position coordinates are calculated at the detection time (time t2 to t5 in FIG. 38) in the interpolated portion. .
- the hypothesis generation unit 32f generates a trajectory link candidate / identification information pair using the trajectory link candidates left without being rejected, and generates a hypothesis.
- the generation process of the trajectory link candidate / identification information pair and the hypothesis is the same as in the first embodiment.
- the environment information storage unit 38 stores the location of the obstacle and the index (cost) of the travel time when the moving object passes through the location where the obstacle exists as the environment information.
- the environment information storage unit 38 may store the moving time of the moving body on the passable path in the tracking area as environment information.
- processing of the hypothesis generation unit 32f using such environment information that is, trajectory link candidate rejection determination processing
- the hypothesis generation unit 32f determines the trajectory from the disappearance time of the trajectory and the occurrence time of the next trajectory for the trajectory link candidate including the trajectory break. Calculate the actual travel time of the moving object while is interrupted.
- the hypothesis generation unit 32f refers to the environment information and calculates an estimated travel time on a path that can be passed from the vicinity of the locus disappearance position to the vicinity of the next locus generation position in the locus connection candidate.
- the hypothesis generation unit 32f rejects the focused trajectory link candidate.
- FIG. 39 schematically shows the environmental information set in the moving time of the moving object on the path that can be passed in the tracking area.
- a plurality of nodes n are set in an area where there is no obstacle Ob that becomes an obstacle to a moving object in the tracking area.
- the node n is a point representing a position coordinate where a moving body can exist in a tracking area given with an arbitrary resolution.
- an edge e that connects the nodes n and satisfies a condition that does not intersect with the obstacle Ob is set. This edge e is set as a straight line.
- nine nodes n are set, and node identification numbers “1” to “9” are assigned.
- an edge that does not intersect with the obstacle Ob such as an edge e that connects the nodes 1 and 2 and an edge e that connects the nodes 2 and 5, is set as the edge e.
- a moving time is set when the moving body moves along the edge.
- the travel time set for each edge is referred to as cost.
- costs are listed in the vicinity of each edge e. For example, the moving time (cost) when the moving body moves along the edge e between the nodes 1 and 4 is set to “1”.
- the moving time (cost) for the moving body to move along the edge e between the nodes 1 and 2 is set to “1.5”.
- the position of the node n and the cost of the edge e are set in advance as environment information indicating the travel time on a route through which the mobile body can pass, and are stored in the environment information storage unit 38.
- This environment information can be said to be a route graph in which costs are associated with edges.
- the hypothesis generation unit 32f interpolates from the vicinity of the locus disappearance position to the vicinity of the next locus generation position with the edge in the locus connection candidate, and generates the next locus from the vicinity of the locus disappearance position with the total cost of the interpolated edge. Calculated as the estimated travel time to the vicinity of the position. If the estimated moving time (total cost of the interpolated edge) is larger than the actual moving time, the hypothesis generating unit 32f rejects the focused trajectory link candidate.
- the hypothesis generation unit 32f calculates the distance between the disappearance position coordinates of the locus a and each node, and selects the node closest to the disappearance position coordinates of the locus a. Similarly, the hypothesis generation unit 32f calculates the distance between the next generation position coordinate of the locus b and each node, and selects the node closest to the generation position coordinate of the locus b. In FIG. 39, the hypothesis generation unit 32f selects the node 4 closest to the disappearance position coordinate of the trajectory a and selects the node 5 closest to the generation position coordinate of the trajectory b.
- the hypothesis generation unit 32f searches for a route on the graph connecting the two selected nodes 4 and 5, and calculates a sum of costs corresponding to each edge included in the searched route. Then, the hypothesis generation unit 32f determines that the total cost of the edges included in the searched route (that is, the estimated movement time) is based on the actual movement time (that is, the time between the disappearance detection time of the locus a and the occurrence detection time of the locus b). If it is also larger, the trajectory link candidate of interest is rejected.
- the hypothesis generation unit 32f treats each path between the two nodes 4 and 5 (that is, a combination of interpolated edges) as a different trajectory link candidate, and the total cost of the edge interpolated for each trajectory link candidate. And the actual travel time are compared. Then, trajectory link candidates whose total cost of the interpolated edge is larger than the actual movement time are rejected, while trajectory link candidates whose total cost of the interpolated edge is less than or equal to the actual movement time are rejected. -Used for generating identification information pairs and hypotheses.
- a “first route” of nodes 4 (R) 1 (R) 2 (R) 5, nodes 4 (R) 7 (R) 8 (R ) 5 and a “third route” of nodes 4 (R) 1 (R) 2 (R) 3 (R) 6 (R) 9 (R) 8 (R) 5 are assumed.
- the first route to the third route are considered here.
- the total cost of the third route is “8.5”.
- the hypothesis generation unit 32f rejects the trajectory link candidate including the trajectories a and b.
- the hypothesis generation unit 32f interpolates the first route on the trajectories a and b.
- the trajectory link candidate and the trajectory link candidate obtained by interpolating the second path between the trajectories a and b are employed for generating a trajectory link candidate / identification information pair and a hypothesis.
- the hypothesis generation unit 32f performs various detection times that occur within the actual movement time from the disappearance detection time of the locus a to the occurrence detection time of the locus b (that is, the detection time of the position information and the detection time of the identification information). Calculate the position coordinates on the path at. Since the total cost of the third route is longer than the actual travel time, the trajectory link candidate in which the third route is interpolated between the trajectories a and b is rejected.
- an upper limit is set for the cost of the path to be interpolated, and when the total cost of each path exceeds the upper limit, it is not used for the interpolating process between the trajectories.
- the hypothesis generation unit 32f uses an obstacle map (see FIG. 37) as environment information, or searches / rejects trajectory link candidates using a route graph (see FIG. 39) as environment information. For this reason, the hypothesis generation unit 32f may determine whether or not to reject the trajectory link candidate using any of the environment information, but even if the trajectory link candidate is determined to be rejected using both of the environment information. Good. Alternatively, the hypothesis generation unit 32f may perform the rejection determination of the trajectory link candidate using environment information in a mode different from the obstacle map or the route graph.
- trajectory link candidate generation unit 31, the hypothesis generation unit 32f, and the likelihood calculation unit 33 are realized by, for example, a CPU of a computer that operates according to a moving body trajectory identification program.
- the program storage unit (not shown) of the computer stores the moving body trajectory identification program, and the CPU reads the program, and the functions of the trajectory link candidate generation unit 31, the hypothesis generation unit 32f, and the likelihood calculation unit 33 are performed. Realize.
- FIG. 40 is a flowchart illustrating the process of the moving object trajectory identification device 3f according to the seventh embodiment.
- the same steps as those in the first embodiment are denoted by the same reference numerals, and the description thereof is omitted.
- step S5 processing until the table holding the identification information is updated (that is, step S1 to step S5) is the same as that of the first embodiment.
- the hypothesis generation unit 32f reads the environment information from the environment information storage unit 38, and determines whether to generate a trajectory link candidate / identification information pair and a hypothesis for the trajectory link candidate where the trajectory is interrupted. (Step S71). That is, the hypothesis generation unit 32f determines whether or not the trajectory link candidate in which the trajectory is interrupted is used for hypothesis generation. Details of step S71 are as described above.
- the hypothesis generation unit 32f After step S71, the hypothesis generation unit 32f generates a trajectory link candidate / identification information pair using the trajectory link candidates remaining without being rejected (step S6).
- the generation method of the trajectory link candidate / identification information pair in the seventh embodiment is the same as that in the first embodiment.
- the hypothesis generation unit 32f generates a hypothesis group using the trajectory link candidate / identification information pair (step S7).
- the hypothesis group generation method in the seventh embodiment is the same as that in the first embodiment.
- the processing procedure after hypothesis generation is the same as the processing after step S8 shown in FIG.
- the seventh embodiment by using the environment information stored in the environment information storage unit 38, it is possible to accurately execute the interpolation process of the trajectory link candidate.
- the position coordinates of the interpolated portion can be accurately specified, so that the identification information likelihood can be calculated more accurately.
- the total number of hypotheses generated by the hypothesis generation unit 32f can be reduced. Therefore, the likelihood calculation process is speeded up.
- the modification of the first embodiment may be applied to the seventh embodiment, and the likelihood calculating unit 33 may not include the trajectory link likelihood calculating unit 331 and may not calculate the trajectory link likelihood.
- the identification information likelihood calculation unit 332 estimates the hypothesis having the highest identification information likelihood as the maximum likelihood hypothesis.
- the likelihood calculation unit 33 does not include the trajectory link likelihood calculation unit 331, and the mobile trajectory identification device 3f includes the trajectory link likelihood calculation unit 331a as in the second embodiment. It is good.
- the identification information likelihood calculation unit 332 estimates the hypothesis having the highest identification information likelihood as the maximum likelihood hypothesis. According to this configuration, the same effect as in the second embodiment can be obtained.
- the likelihood calculation unit 33 is combined with the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, and the identification information likelihood calculation unit 332 in combination with the seventh embodiment and the fifth embodiment, and the attribute information likelihood.
- a calculation unit 334 may be included.
- the moving body trajectory identification system also includes an attribute information detection device 5.
- the likelihood calculation unit 33 is combined with the likelihood calculation control unit 330, the trajectory link likelihood calculation unit 331, and the identification information likelihood calculation unit 332 in combination with the seventh embodiment and the sixth embodiment, and the movement information likelihood.
- a calculation unit 335 may be provided.
- the moving body trajectory identification system also includes a movement information detection device 6.
- FIG. 41 is a block diagram showing the minimum configuration of the moving object trajectory identification system of the present invention.
- 41 includes a trajectory link candidate generation unit 91, a hypothesis generation unit 92, and a likelihood calculation unit 93.
- the trajectory link candidate generation unit 91 (corresponding to the trajectory link candidate generation units 31 and 31c) generates a trajectory link candidate that is a combination of trajectories detected in a past fixed time.
- the hypothesis generation unit 92 (corresponding to the hypothesis generation units 32 and 32c) combines the trajectory link candidate generated by the trajectory link candidate generation unit 91 and the identification information detected in the past fixed time, and combines the trajectory link candidate / identification.
- a set of information pairs is generated, and a hypothesis that is a set of trajectory link candidate / identification information pairs that satisfies a predetermined condition (corresponding to the first condition to the third condition described above) is generated.
- the likelihood calculation unit 93 selects individual hypotheses and represents the trajectory link candidate for each trajectory link candidate / identification information pair belonging to the selected hypothesis. An identification information likelihood that is the likelihood that the identification information is detected for the trajectory is calculated. Then, the identification information likelihood is calculated for the selected hypothesis by integrating the identification information likelihood, and the maximum likelihood hypothesis is estimated based on the calculated identification information likelihood.
- the detected trajectory is the trajectory of which mobile object. Whether it is present or not can be determined with high accuracy.
- a trajectory link candidate generation unit (31, 31c) that generates a trajectory link candidate that is a combination of trajectories detected in a past fixed time, a trajectory link candidate generated by the trajectory link candidate generation unit, and a predetermined past time
- a predetermined condition first condition to third condition
- the likelihood of calculating the identification information likelihood, calculating the identification information likelihood for the selected hypothesis by integrating the identification information likelihood for each trajectory link candidate / identification information pair, and estimating the maximum likelihood hypothesis based thereon Moving locus identification system comprising a calculation unit (33,33a, 33b, 33c).
- the mobile object trajectory identification system further includes a map storage unit (37) for storing a probability map that defines the detection probability of the identification information corresponding to each position in the tracking area of the mobile object.
- the likelihood calculation unit selects the individual hypothesis, the likelihood calculation control unit (330), and the position of the moving object for each detection time of the identification information in the trajectory link candidate / identification information pair belonging to the selected hypothesis.
- the detection probability at that position is identified from the probability map, and the identification information likelihood for the trajectory link candidate / identification information pair is calculated using the detection probability, and the identification information likelihood Are integrated, and an identification information likelihood calculation unit (332, 332b, 332c) that calculates the identification information likelihood of the selected hypothesis, and a maximum likelihood estimation that estimates the maximum likelihood hypothesis based on the identification information likelihood of the hypothesis (I.e., identification information likelihood calculators 332, 332b, 332c, attribute information likelihood calculator 334, or movement information likelihood calculator 335).
- the moving body trajectory identification system further includes a likelihood storage unit (34) for storing a trajectory / identification information correspondence likelihood that is a likelihood that the trajectory and the identification information are associated with each other.
- a likelihood storage unit (34) for storing a trajectory / identification information correspondence likelihood that is a likelihood that the trajectory and the identification information are associated with each other.
- the identification information likelihood calculation unit (332b) estimates the maximum likelihood hypothesis, the latest likelihood included in the trajectory link candidate in the trajectory link candidate / identification information pair belonging to all hypotheses generated by the hypothesis generation unit together with the maximum likelihood hypothesis.
- a trajectory / identification information correspondence likelihood is derived for a set of a trajectory and identification information, and the trajectory / identification information correspondence likelihood is stored in a likelihood storage unit.
- the likelihood corresponding to the trajectory / identification information corresponding to the latest combination of the trajectory and the identification information and the detection probability specified for each detection time of the identification information Based on this, the identification information likelihood for the trajectory link candidate / identification information pair is calculated.
- the accuracy of the association between the locus and the identification information is further improved. Can do.
- the moving body trajectory identification system includes a hypothesis storage unit (333) that stores all hypotheses generated by the hypothesis generation unit together with the maximum likelihood hypothesis, and whether or not there is a change in identification information detected in the past certain time.
- An information history change determination unit (35) for determining
- the trajectory link candidate generation unit generates a trajectory link candidate
- the hypothesis generation unit (32c) generates a hypothesis group
- each hypothesis is determined. select.
- the information history change determination unit determines that there is no change in the acquired information
- individual hypotheses are selected from the hypothesis group stored in the hypothesis storage unit.
- the maximum likelihood hypothesis is estimated from the hypothesis group generated by the hypothesis generation unit, and all hypotheses are stored in the hypothesis storage unit together with the maximum likelihood hypothesis.
- the hypothesis generation unit repeatedly generates the same hypothesis group as the previously generated hypothesis group, duplicate generation of the same hypothesis group is not performed. The amount can be reduced.
- the hypothesis generation unit associates the attribute information of the moving object detected in the past certain time with the trajectory link candidate, the trajectory link candidate associated with the attribute information, and the past certain time
- a hypothesis group that is a set of trajectory link candidate / identification information pairs combined with the detected identification information and that satisfies a predetermined condition is generated.
- the likelihood calculation unit is a mobile object in which each attribute information associated with the trajectory link candidate is the same for each trajectory link candidate of the trajectory link candidate / identification information pair belonging to each hypothesis selected from the hypothesis group. Attribute information that calculates the attribute information likelihood for each selected hypothesis by calculating the attribute information likelihood that is the likelihood that represents and integrating the attribute information likelihood for each trajectory link candidate / identification information pair
- a likelihood calculation unit (334) is provided.
- the maximum likelihood estimation unit estimates the maximum likelihood hypothesis based on the identification information likelihood and the attribute information likelihood of the hypothesis.
- the trajectory link combining the trajectory link candidate associated with the moving information of the moving body detected by the hypothesis generation unit in the past certain time and the identification information detected in the past certain time.
- a hypothesis group that is a set of candidate / identification information pairs and that satisfies a predetermined condition is generated.
- the likelihood calculation unit for each trajectory link candidate of the trajectory link candidate / identification information pair belonging to each hypothesis selected from the hypothesis group, from the movement information associated with the trajectory link candidate and the trajectory link candidate
- the movement information likelihood representing the likelihood that the derived movement information is for the same moving object is calculated, and the movement information likelihood is integrated for each trajectory link candidate / identification information pair, and thus selected.
- a movement information likelihood calculation unit (335) for calculating the movement information likelihood related to the hypothesis is provided. Further, the maximum likelihood estimation unit estimates the maximum likelihood hypothesis based on the identification information likelihood and the movement information likelihood of the hypothesis.
- the likelihood calculation unit connects the trajectory included in the trajectory link candidate for each trajectory link candidate of the trajectory link candidate / identification information pair belonging to each hypothesis selected from the hypothesis group generated by the hypothesis generation unit.
- a trajectory link likelihood calculation unit (331) for calculating a trajectory link likelihood indicating likelihood is provided.
- the maximum likelihood estimation unit estimates the maximum likelihood hypothesis based on the identification information likelihood and the trajectory link likelihood of the hypothesis. According to the above configuration, since the maximum likelihood hypothesis is estimated in consideration of not only the identification information likelihood but also the trajectory link likelihood, it is possible to further improve the accuracy of association between the trajectory and the identification information.
- the trajectory link likelihood calculation unit calculates the trajectory link likelihood based on the position coordinates detected for the moving body that defines the trajectory and the detection time.
- the trajectory link likelihood calculating unit (331a) calculates the trajectory link likelihood that is the likelihood of connecting the trajectories included in the trajectory link candidate for each trajectory link candidate generated by the trajectory link candidate generation unit.
- a trajectory link candidate having a trajectory link likelihood equal to or greater than a threshold is selected.
- the hypothesis generation unit generates a hypothesis using the selected trajectory link candidate. According to the above configuration, since the total number of hypotheses generated by the hypothesis generation unit can be reduced, the association between the trajectory and the identification information can be performed at high speed.
- the moving object trajectory identification system stores, as environmental information, the location of an obstacle existing in the tracking area of the moving object and an indicator of the moving time when the moving object passes through the place where the obstacle exists.
- An environmental information storage unit (338) is provided.
- the hypothesis generation unit calculates the actual movement time from the disappearance detection time of the locus in the locus connection candidate to the occurrence detection time of the next locus, and the next locus appears after the locus disappears based on the environment information.
- the moving time of the moving object is estimated until the time.
- a trajectory link candidate whose estimated travel time is longer than the actual travel time is excluded from the trajectory link candidates used for hypothesis generation.
- the environment information storage unit stores environment information indicating a travel time on a route through which the mobile body can pass through the tracking area.
- the hypothesis generation unit calculates the actual movement time from the disappearance detection time of the locus to the occurrence detection time of the next locus for the locus connection candidate, and after the locus disappears based on the environment information, Estimate the moving time of the moving object until it appears.
- a trajectory link candidate whose estimated travel time is longer than the actual travel time is excluded from the trajectory link candidates used for hypothesis generation.
- the present invention is applied to a moving object trajectory identification technique for identifying which moving object is a trajectory detected in a tracking area. For example, by tracking the position of a person working in an office or factory in association with each employee's unique employee number, it is possible to determine whether the area is accessible according to the security authority for each person. It can be applied as a security technology, such as realizing an alert function when a person enters a prohibited area. In addition, the present invention can be applied to marketing techniques such as measuring the trends of shoppers by tracking the positions of persons shopping in a shopping center and the member numbers unique to each person in association with each other.
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Radar Systems Or Details Thereof (AREA)
- Burglar Alarm Systems (AREA)
- Traffic Control Systems (AREA)
Abstract
Description
本願は、日本国に出願された特願2009-191358号(2009年8月20日出願)及び特願2009-280926号(2009年12月10日出願)に基づき優先権を主張し、その内容をここに援用する。
尚、軌跡連結候補と識別情報との組を「軌跡連結候補・識別情報ペア」と記す。また、「unknown」を含む組も軌跡連結候補・識別情報ペアに該当する。
第1の条件は、仮説に属する軌跡連結候補・識別情報ペアの間で軌跡の重複及び識別情報の重複が生じないことである。例えば、4つの軌跡k1、k2、k3、k4について識別情報ID1、ID2が検出されている場合、[{(k1,k2),ID1},{(k1,k3,k4)、ID2}]という集合では、軌跡k1が重複しているので、第1の条件を満たさない。[{(k1,k2),ID1},{(k3,k4)、ID1}]という集合では、識別情報ID1が重複しているので第1の条件を満たさない。一方、[{(k1,k2),ID1},{(k3,k4)、ID2}]という集合では、軌跡及び識別情報のいずれも重複していないので、第1の条件を満たす。ここで、(k1,k2)は軌跡連結候補を表し、{(k1,k2)、ID1}は軌跡連結候補・識別情報ペアを表す。また、仮説において1つの移動体の軌跡は、1つの軌跡連結候補に纏められることを前提とする。仮説は、各軌跡がどの移動体の軌跡であるかを決定するために用いる候補であり、軌跡や識別情報の重複を含む仮説は適切な候補とは言えないため、第1の条件を設けた。
実施例1において、尤度計算部33は仮説毎に2種類の尤度を計算する。第1の尤度は、仮説に属する軌跡連結候補・識別情報ペアにおける軌跡連結候補に含まれる軌跡の連結の尤もらしさである。第1の尤度を軌跡連結尤度と記す。第2の尤度は、軌跡連結候補・識別情報ペアにおける軌跡連結候補が表す起動について識別情報が検出される尤もらしさである。第2の尤度を識別情報尤度、或いは識別情報検出尤度と記す。軌跡連結候補が表す軌跡をSとし、識別情報をOとすると、識別情報尤度は条件付き確率P(OiS)として表される。
尚、識別情報尤度は軌跡連結候補・識別情報ペア毎に算出され、仮説に属する軌跡連結候補・識別情報ペアの識別情報尤度を統合した結果が当該仮説の識別情報尤度となる。
ここで、軌跡連結候補の組み合わせにおいて、同時刻に重複して検出された軌跡を組み合わせる場合と、異なる時刻に独立に検出された軌跡を組み合わせる場合がある。
尚、尤度計算部33において軌跡連結尤度の閾値を予め定めておき、その閾値未満の軌跡連結尤度について、以降、処理対象から除外するようにしてもよい。
次に、軌跡連結尤度計算部331は順番が隣り合う2つの軌跡の夫々について、軌跡連結尤度を計算する。即ち、軌跡連結尤度計算部331は軌跡消失時刻の順番が1番目及び2番目の軌跡について、軌跡連結尤度を計算する。同様に、軌跡消失時刻の順番が2番目及び3番目の軌跡について軌跡連結尤度を計算し、軌跡消失時刻の順番が3番目及び4番目の軌跡について軌跡連結尤度を計算する。尚、順番が隣り合う2つの軌跡に係る軌跡連結尤度の計算については前述の方法を用いるものとする。
識別情報尤度計算部332は、予め識別情報検出装置2毎に定められた識別情報確率マップを参照して識別情報尤度を計算する。識別情報確率マップとは、追跡領域50をグリッド分割し、分割した各セル内において識別情報が識別情報検出装置2により検出される確率を0~1の間の数値で定義したマップである。
先ず、識別情報尤度計算部332は仮説に含まれる軌跡連結候補・識別情報ペアを夫々抽出する。そして、識別情報尤度計算部332は軌跡連結候補・識別情報ペア毎に識別情報尤度を計算して積算する。各仮説に含まれる軌跡連結候補・識別情報ペアの数は一定ではなく、識別情報尤度が算出されない軌跡連結候補・識別情報ペアも存在する。そのため、識別情報尤度計算部332は軌跡連結候補・識別情報ペア毎に計算した識別情報尤度の積算結果を正規化する。
尚、識別情報尤度計算部332は識別情報尤度を計算できない軌跡連結候補・識別情報ペアについては識別情報尤度を計算しない。即ち、識別情報又は軌跡連結候補が「unknown」となっている軌跡連結候補・識別情報ペアについては識別情報尤度を計算しない。
また、識別情報尤度計算部332は仮説毎に軌跡連結尤度及び識別情報尤度を統合し、以って、仮説全体の尤度を計算する。尤も高い尤度を有する仮説が、軌跡と識別情報との関係を最適に表していると言える。識別情報尤度計算部332は、最も高い尤度を有する仮説を選択して、識別結果出力装置4に出力する。
図6は、追跡領域50内を移動する移動体A及び移動体Bについて時刻t1~t9における位置及び軌跡番号を示す。ここでは、移動体Aの軌跡が軌跡1及び軌跡3に相当し、移動体Bの軌跡が軌跡2及び軌跡4に相当するものとする。また、移動体Aのみの識別情報が検出可能であり、移動体Bの識別情報の登録を行なっておらず、対応する識別情報が存在しないものとする。
図11、図12、図13、及び図14は追跡領域の識別情報確率マップ上での識別情報の検出確率値の取得手順を説明する図である。
先ず、軌跡連結候補生成部31は、位置情報検出装置1から軌跡番号、検出時刻、及び位置座標の組を取得する(ステップS1)。図6に示すように複数の座標を時刻順に検出した場合、位置情報検出装置1は軌跡番号、検出時刻、及び位置座標の組として、例えば時刻t1について、{1,t1,p(30,80)}と{2,t1,p(25,10)}の組を移動体軌跡識別装置3の軌跡連結候補生成部31に出力する。時刻t1以降の時刻t2、t3、・・・についても軌跡番号、検出時刻、及び位置座標の組が軌跡連結候補生成部31に提供される。
また、軌跡連結候補生成部31は過去一定時間における軌跡番号、検出時刻、及び位置座標の組(即ち、ステップS2の更新処理で得られる軌跡番号、検出時刻、及び位置座標の組)を自身で記憶するだけでなく、尤度計算部33に出力する。異なる軌跡の位置座標であって同時刻に検出された位置情報の重心点を計算している場合には、その重心点も尤度計算部33に出力する。軌跡連結候補生成部31は、これらの情報を仮説生成部32を介して尤度計算部33に出力する。
また、仮説生成部32はステップS5で更新したテーブルに格納されている識別情報と、ステップS3軌跡連結候補生成部31から入力された軌跡連結候補を用いて、両者の組、即ち軌跡連結候補・識別情報ペアを生成する(ステップS6)。図6及び図10に示す例では、仮説生成部32は{(軌跡1,軌跡3),ID1}、{(軌跡2,軌跡4),ID1}、・・・の軌跡連結候補・識別情報ペアを生成する。ここでは、軌跡連結候補に2つの軌跡を含むものとしたが、軌跡連結候補に含まれる軌跡の数は2に限定されない。また、仮説生成部32は、ステップS6において、対応する相手が存在しないことを示す「unknown」を含む軌跡連結候補・識別情報ペアも生成する。従って、{(軌跡1,軌跡3),unknown}、{(軌跡2,軌跡4),unknown}、{(軌跡1,軌跡4),unknown}などの軌跡連結候補・識別情報ペアも生成する。
図6に示す軌跡1と軌跡3の軌跡連結尤度を計算する場合、軌跡1の消失時刻t4と軌跡3の出現時刻t6、並びに軌跡1の消失位置座標p(90,60)と軌跡3の出現位置座標p(130,55)に基づき、軌跡1の消失から軌跡3の出現までの移動速度を計算する。この移動速度と予め定義した移動速度との誤差を計算し、当該誤差をパラメータとする尤度関数から軌跡連結尤度を取得する。尚、軌跡連結尤度の計算方法はこの方法に限定されるものではなく、他の方法であってもよい。
従来の移動体軌跡識別方法では、図1乃至図14に示すように、時刻t4~t6で追跡の途切れが生じると、その後検出される軌跡3、軌跡4については、時刻t8で新たに識別情報を検出するまで軌跡と識別情報との対応付けを行なうことができない。また、図11乃至図14において、軌跡3、軌跡4は時刻t8、t9において夫々同一セルに位置する。この場合、時刻t8で検出した識別情報を軌跡3、軌跡4のいずれに対応付けるのかを判別することが困難となる。
識別装置尤度計算部332は、実施例1と同様に、仮説毎に軌跡連結候補・識別情報ペアに対する識別情報尤度を計算する。但し、実施例2では、識別情報尤度計算部332は識別情報尤度が最も高い仮説を識別結果出力装置4に出力する。実施例2では、識別情報尤度計算部332は軌跡連結候補と識別情報尤度の統合結果ではなく、識別情報尤度に基づいて識別結果出力装置4に送出すべき仮説を判定する。
図17及び図18は、実施例2の移動体軌跡識別装置3aの処理を示すフローチャートである。先ず、軌跡連結候補生成部31は位置情報検出装置2から軌跡番号、検出時刻、及び位置座標の組を取得する(ステップS21)。
実施例2では、軌跡連結尤度計算部331aが軌跡連結候補の軌跡連結尤度を計算し、軌跡連結尤度が閾値未満となっている軌跡連結候補を仮説生成部32に出力しないことにより、その軌跡連結候補が仮説生成に使用されないよう予め除去する。従って、仮説生成部32において軌跡連結候補・識別情報ペアを生成する処理(ステップS27)及び仮説を生成する処理(ステップS28)の仕事量を低減できる。その結果、軌跡と識別情報との対応付けを高速に行なうことができる。即ち、どの軌跡がどの移動体の軌跡であるかを高速に判定することができる。
図21及び図22は、実施例3に係る移動体軌跡識別装置3bの処理を示すフローチャートである。ここで、実施例1の移動体軌跡識別装置3と同様の処理については、図4及び図5に示すフローチャートと同一の符号を付し、その説明を省略する。図21のステップS1乃至ステップS7により仮説群を生成する処理は、図4のステップS1乃至ステップS7と同様である。但し、ステップS7において、実施例1において述べた第1の条件乃至第3の条件を全て満たす仮説を生成してもよい。何故なら、第3の条件を満たさない仮説では識別装置尤度を計算しないため仮説の尤度を計算できないが、実施例3では尤度記憶部34に記憶されている軌跡・識別情報対応確率を参照することにより、仮説の尤度を算出できるからである。
ステップS11の後、尤度計算部33bはステップS8乃至ステップS11を繰り返し実行する。ステップS8において、尤度未計算の仮説が存在しないと判定された場合(即ち、ステップS8の判定結果「No」)、識別情報尤度計算部332bは全仮説内の軌跡連結候補・識別情報ペアに基づいて今回の尤度計算処理(ステップS8乃至ステップS11のループ)の結果を反映した軌跡・識別情報対応確率を尤度記憶部34に記憶する(ステップS11b)。識別情報尤度計算部332bは、軌跡連結候補・識別情報ペアにおける識別情報と、その軌跡連結候補における最新の軌跡の組に対応付けテレビジョン受像機、当該軌跡連結候補・識別情報ペアの識別情報尤度を軌跡・識別情報対応確率として尤度記憶部34に記憶する。但し、最新の軌跡及び識別情報の組が共通する軌跡連結候補・識別情報ペアが複数の仮説に含まれている場合を考慮する必要がある。例えば、或る仮説に{(軌跡1,軌跡4),ID1}が含まれ、別の仮説に{(軌跡2,軌跡4),ID1}が含まれ、さらに別の仮説に{(軌跡3,軌跡4),ID1}が含まれており、軌跡4が最新の軌跡である場合を想定する。この3つの軌跡連結候補・識別情報ペアはいずれも最新の軌跡及び識別情報の組が「軌跡4、ID1」となる。このように、最新の軌跡及び識別情報の組が共通する軌跡連結候補・識別情報ペアが複数存在する場合、その複数の軌跡連結候補・識別情報ペアについて計算された識別情報尤度を加算して、その加算結果を最新の軌跡及び識別情報に関する軌跡・識別情報対応確率として記憶してもよい。
ステップS11bの後、識別情報尤度計算部332bはステップS11で計算した尤度統合値が最大となる最尤仮説を推定して、識別結果出力装置4に送出する(ステップS12)。識別結果出力装置4は、最尤仮説に基づいて、どの軌跡がどの移動体の軌跡であるかを判断して出力する(ステップS13)。尚、実施例3におけるステップS12、S13は実施例1と同様である。
図26及び図27は、実施例4に係る移動体軌跡識別装置3cの処理を示すフローチャートである。ここで、実施例1と同様の処理(図4及び図5参照)には同一の符号を付し、その説明を省略する。
図30及び図31は、実施例5に係る移動体軌跡識別装置3dの処理を示すフローチャートである。ここで、実施例1と同一のステップ(図4及び図5参照)には同一の符号を付し、その説明を省略する。図30において、取得した識別情報を保持するテーブルを更新するまでの処理(即ち、ステップS1乃至ステップS5)は、実施例1(図4参照)と同様である。
尚、仮説生成部32dは、更新後のテーブルを自身で記憶するだけでなく、尤度計算部33dに出力する。
このように、移動情報検出時における移動体の位置座標は、移動情報検出装置6の種類(換言すれば、移動情報種別)に応じた方法で求めればよい。
先ず、仮説生成部32eは軌跡連結候補生成部31から入力した軌跡連結候補と、現時刻から過去一定時間遡った時刻までの時間帯に移動情報検出装置6により検出された移動情報を用いて、軌跡連結候補と移動情報の組(即ち、移動情報付き軌跡連結候補)を生成する。
先ず、仮説生成部32eは移動情報検出装置6から入力した移動情報のうち、位置座標を有する移動情報を軌跡連結候補と対応付ける。仮説生成部32eは、移動情報検出装置6から入力した移動情報から位置座標を有する移動情報のみを選別し、その中から1つの移動情報を選択する。そして、仮説生成部32eは選択した移動情報の検出時刻における軌跡連結候補の位置座標と選択した移動情報の位置座標とのユークリッド距離を全ての軌跡連結候補と選択した移動情報の組について計算する。次に、仮説生成部32eはユークリッド距離が最短となる軌跡連結候補と移動情報の組を特定して記憶する。この処理を移動情報検出装置6から入力した全ての移動情報に対して1回ずつ実施する。この処理によって移動情報が対応付けられた軌跡連結候補を「仮の移動情報付き軌跡連結候補」と記す。
移動情報尤度計算部335は、仮説生成部32eにより生成された仮説を入力し、各仮説について移動情報尤度を計算する。移動情報尤度計算部335は、個々の移動情報尤度を計算する場合、仮説に含まれる個々の軌跡連結候補毎に移動情報尤度を計算し、同じ仮説に属する軌跡連結候補毎に計算した移動情報尤度を積算して正規化し、以って、仮説の移動情報尤度を算出する。
上記の例では、コサイン類似度を計算して移動方向ベクトルの類似度を計算して移動情報尤度としたが、他の計算方法で移動方向ベクトルの類似度を計算してもよい。
先ず、移動情報尤度計算部335は軌跡連結候補から計算した速度をいずれかの速度区分に分類する。各速度区分における速度範囲は予め設定される。移動情報尤度計算部335は、移動情報が示す速度区分と、軌跡連結候補から計算した速度に基づいて判断した速度区分との組み合わせに基づき、移動情報尤度を決定する。この場合、速度区分の組み合せ毎に移動情報尤度を予め定義して、移動情報尤度計算部335に保持する。即ち、[静止、静止]、[静止、歩行]、[静止、走行]、[歩行、歩行]、[歩行、走行]、[走行、走行]のような移動情報の速度区分と、軌跡連結候補から計算した速度区分の組み合わせの全通りについて、移動情報尤度を予め定義して、移動情報尤度計算部335に保持する。このとき、速度区分の内容が近似するほど移動情報尤度を高く設定し、速度区分の内容が異なるほど移動情報尤度を低く設定する。例えば、[静止、静止]、[歩行、歩行]、[走行、走行]などの同一速度区分の組み合わせでは、移動情報尤度を高く設定し、[静止、走行]のように速度が大きく異なる場合には、移動情報尤度を低く設定する。また、[静止、歩行]や[歩行、走行]のように速度区分が同一ではないが、[静止、走行]ほどの違いはない組み合わせについては、中程度の移動情報尤度を定義すればよい。そして、移動情報尤度計算部335は移動情報が示す速度区分と軌跡連結候補から計算した速度区分との組み合わせに対応する移動情報尤度を予め定義した移動情報尤度の中から選択する。
図34及び図35は、実施例6に係る移動体軌跡識別装置3eの処理を示すフローチャートである。ここで、実施例1と同一のステップには同一の符号を付し、その説明を省略する。また、図34において識別装置を保持するテーブルを更新するまでの処理(即ち、ステップS1乃至ステップS5)は、実施例1と同様である。
また、仮説生成部32eは、更新後のテーブルを自身で記憶するだけでなく、尤度計算部33eに出力する。
次に、仮説生成部32fは軌跡aの消失位置座標(Xa,Xb)から軌跡bの発生位置座標(Xb,Yb)の間に直線を内挿し、その直線上の位置座標であり、かつ、実移動時間内における検出時刻t2乃至t5での位置座標を計算する。尚、時刻t2、t3、t4、t5における位置に該当するセルのコストは、「0」、「0.5」、「3.0」、「0.5」であるため、内挿した区間におけるコストの総和は「0+0.5+3.0+0.5=4.0」となる。
図40は、実施例7に係る移動体軌跡識別装置3fの処理を示すフローチャートである。ここで、実施例1と同一のステップには同一の符号を付し、その説明を省略する。
ステップS5の後、仮説生成部32fは環境情報記憶部38から環境情報を読み込み、軌跡の途切れが生じている軌跡連結候補について、軌跡連結候補・識別情報ペア及び仮説を生成するか否かを判定する(ステップS71)。即ち、仮説生成部32fは軌跡の途切れが生じている軌跡連結候補を仮説生成に用いるか否かを判定する。ステップS71の詳細については、前述した通りである。
仮説生成部92(仮説生成部32、32cに相当)は、軌跡連結候補生成部91により生成された軌跡連結候補と、過去一定時間に検出された識別情報とを組み合わせて、軌跡連結候補・識別情報ペアの集合を生成し、所定の条件(前述の第1条件乃至第3条件に相当)を満たす軌跡連結候補・識別情報ペアの集合である仮説を生成する。
(1)過去一定時間に検出された軌跡の組み合わせである軌跡連結候補を生成する軌跡連結候補生成部(31、31c)と、軌跡連結候補生成部で生成された軌跡連結候補と、過去一定時間に検出された移動体の識別情報とを組み合わせた軌跡連結候補・識別情報ペアの集合であって、所定の条件(第1の条件乃至第3の条件)を満たす集合である仮説を生成する仮説生成部(32、32c)と、個々の仮説を選択し、選択した仮説に属する軌跡連結候補・識別情報ペア毎に、その軌跡連結候補が表す軌跡についてその識別情報が検出される尤もらしさである識別情報尤度を計算し、軌跡連結候補・識別情報ペア毎に識別情報尤度を統合することによって、選択した仮説に対する識別情報尤度を算出し、それに基づいて最尤仮説を推定する尤度計算部(33、33a、33b、33c)を具備する移動体軌跡識別システム。
上記の構成では、軌跡・識別情報対応尤度により過去一定時間より以前の識別情報も識別情報尤度の計算に反映させることができるので、軌跡と識別情報との対応付けの精度をより高めることができる。
上記の構成では、仮説生成部が過去に生成した仮説群と同じ仮説群を繰り返して生成するような場合に、同一仮説群の重複生成を行わないので、軌跡と識別情報との対応付けにおける計算量を低減することができる。
上記の構成によれば、識別情報尤度だけでなく軌跡連結尤度も考慮して最尤仮説を推定するので、軌跡と識別情報との対応付けの精度をより高めることができる。
上記の構成によれば、仮説生成部が生成する仮説の総数を少なくできるので、軌跡と識別情報との対応付けを高速に行うことができる。
例えば、オフィスや工場で勤務する人物の位置と各人物固有の社員番号とを対応付けて追跡することにより、人物毎のセキュリティ権限に応じて立ち入り可能なエリアであるか否かを判別し、不審者が立入り禁止エリアに入った際のアラート機能を実現するなどの、セキュリティ技術として適用可能である。
また、ショッピングセンター内で買い物をする人物の位置と各人物固有の会員番号とを対応付けて追跡することにより、買い物客の動向を計測するなどのマーケティング技術にも適用可能である。
2 識別情報検出装置
3 移動体軌跡識別装置
4 識別結果出力装置
5 属性情報検出装置
6 移動情報検出装置
31 軌跡連結候補生成部
32 仮説生成部
33 尤度計算部
34 尤度記憶部
37 マップ記憶部
38 環境情報記憶部
50 追跡領域
91 軌跡連結候補生成部
92 仮説生成部
93 尤度計算部
330 尤度計算制御部
331 軌跡連結尤度計算部
332 識別装置尤度計算部
333 仮説記憶部
334 属性情報尤度計算部
335 移動情報尤度計算部
Claims (33)
- 過去一定時間に検出された移動体の軌跡の組み合わせである軌跡連結候補を生成する軌跡連結候補生成部と、
過去一定時間に検出された移動体の識別情報を軌跡連結候補と組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、その軌跡連結候補・識別情報ペアの集合で所定条件を満たすものを仮説として生成する仮説生成部と、
個々の仮説を選択し、選択した仮説に属する軌跡連結候補・識別情報ペアにおいて、その軌跡連結候補が表す軌跡について識別情報が検出される尤もらしさを示す識別情報尤度を計算し、軌跡連結候補・識別情報ペア毎に識別情報尤度を統合することにより、選択した仮説に関する識別情報尤度を算出し、仮説の識別情報尤度に基づいて最尤仮説を推定する尤度計算部を具備する移動体軌跡識別システム。 - 追跡領域内の位置座標に応じた移動体の識別情報の検出確率を予め定義した確率マップを記憶するマップ記憶部を更に具備し、
前記尤度計算部は、
選択された仮説に属する軌跡連結候補・識別情報ペアにおける識別情報の検出時刻毎に、移動体の位置を軌跡連結候補が示す軌跡上に特定し、その位置で移動体が検出される確率値を確率マップから読み取り、その確率値に基づいて識別情報尤度を計算し、更に、軌跡連結候補・識別情報ペア毎に識別情報尤度を統合することにより、選択した仮説に関する識別情報尤度を算出する識別情報尤度計算部と、
仮説の識別情報尤度に基づいて最尤仮説を推定する最尤推定部を具備する請求項1記載の移動体軌跡識別システム。 - 移動体の軌跡と識別情報とが対応付けられる尤もらしさである軌跡・識別情報尤度を記憶する尤度記憶部を更に具備し、
前記識別情報尤度計算部は、最尤仮説とともに仮説生成部により生成された全ての仮説に属する軌跡連結候補・識別情報ペアにおいて、その軌跡連結候補に含まれる最新の軌跡と識別情報との組に応じた軌跡・識別情報対応尤度を導出して尤度記憶部に記憶し、
軌跡連結候補・識別情報ペアにおいて、軌跡連結候補に含まれる最新の軌跡と識別情報との組に応じた軌跡・識別情報対応尤度と、識別情報の検出時刻毎に確率マップから読み出した確率値に基づいて、軌跡連結候補・識別情報ペアに関する識別情報尤度を算出するようにした請求項2記載の移動体軌跡識別システム。 - 最尤仮説とともに仮説生成部に生成された全仮説を記憶する仮説記憶部と、
過去一定時間に検出された識別情報に変化があるか否かを判定する情報履歴変化判定部を更に具備し、
前記軌跡連結候補生成部は、情報履歴変化判定部により識別情報に変化ありと判定された場合に軌跡連結候補を生成し、
仮説生成部は、情報履歴変化判定部により識別情報に変化ありと判定された場合に仮説群を生成し、
情報履歴変化判定部により識別情報に変化ありと判定された場合、仮説生成部が生成した仮説群から個々の仮説を選択し、一方、情報履歴変化判定部により識別情報に変化なしと判定された場合には、仮説記憶部に記憶された仮説群から個々の仮説を選択し
軌跡連結候補生成部は、情報履歴変化判定部により識別情報に変化ありと判定された場合に仮説生成部が生成した仮説群から最尤仮説を推定したとき、その最尤仮説とともに仮説生成部により生成された全仮説を仮説記憶部に記憶するようにした請求項2記載の移動体軌跡識別システム。 - 前記仮説生成部は、過去一定時間に検出された移動体の属性情報を軌跡連結候補に対応付けるとともに、その属性情報が対応付けられた軌跡連結候補に移動体の識別情報を組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、所定条件を満たす軌跡連結候補・識別情報ペアの集合を仮説として生成し、
前記尤度計算部は、選択された仮説に属する軌跡連結候補・識別情報ペアにおいて軌跡連結候補に対応付けられている属性情報が同一の移動体を表す尤もらしさである属性情報尤度を計算し、軌跡連結候補・識別情報ペア毎の属性情報尤度を統合することにより、選択した仮説に関する属性情報を算出する属性情報尤度計算部を具備し、
最尤推定部は、仮説の識別情報尤度及び属性情報尤度に基づいて最尤仮説を推定するようにした請求項2記載の移動体軌跡識別システム。 - 前記仮説生成部は、過去一定時間に検出された移動体の移動情報を軌跡連結候補に対応付けるとともに、その移動情報が対応付けられた軌跡連結候補に移動体の識別情報を組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、所定条件を満たす軌跡連結候補・識別情報ペアの集合を仮説として生成し、
前記尤度計算部は、選択された仮説に属する軌跡連結候補・識別情報ペアにおいて軌跡連結候補に対応付けられている移動情報が同一の移動体を表す尤もらしさである移動情報尤度を計算し、軌跡連結候補・識別情報ペア毎の移動情報尤度を統合することにより、選択した仮説に関する移動情報尤度を算出する移動情報尤度計算部を具備し、
最尤推定部は、仮説の識別情報尤度及び移動情報尤度に基づいて最尤仮説を推定するようにした請求項2記載の移動体軌跡識別システム。 - 前記尤度計算部は、選択された仮説に属する軌跡連結候補・識別情報ペアにおける軌跡連結候補に含まれている軌跡が連結する尤もらしさである軌跡連結尤度を計算する軌跡連結尤度を更に具備し、
最尤推定部は、仮説の識別情報尤度及び軌跡連結尤度に基づいて最尤仮説を推定するようにした請求項2記載の移動体軌跡識別システム。 - 前記軌跡連結尤度は、軌跡を規定する移動体の位置座標とその検出時刻に基づいて軌跡連結尤度を計算するようにした請求項7記載の移動体軌跡識別システム。
- 前記軌跡連結候補生成部は、生成した軌跡連結候補に含まれている軌跡が連結する尤もらしさである軌跡連結尤度を計算し、その軌跡連結尤度が閾値以上となっている軌跡連結候補を選別し、
前記仮説生成部は、選別された軌跡連結候補を用いて仮説群を生成するようにした請求項1記載の移動体軌跡識別システム。 - 追跡領域内に存在する障害物の場所と、その障害物を移動体が通過する際の移動時間とを環境情報として記憶する環境情報記憶部を更に具備し、
前記仮説生成部は、軌跡連結候補における軌跡の消失検出時刻から次の軌跡の出現検出時刻までの実移動時間を計算するとともに、軌跡の消失から次の軌跡の出現までの間に移動体が移動した時間を環境情報に基づいて推定し、その推定移動時間が実移動時間よりも大きくなる軌跡連結候補を仮説生成に用いる軌跡連結候補群から除外するようにした請求項1記載の移動体軌跡識別システム。 - 追跡領域内で移動体が通過可能な経路に沿う移動時間を環境情報として記憶する環境情報記憶部を更に具備し、
前記仮説生成部は、軌跡連結候補における軌跡の消失検出時刻から次の軌跡の出現検出時刻までの実移動時間を計算するとともに、軌跡の消失から次の軌跡の検出までの間に移動体が移動した時間を環境情報に基づいて推定し、その推定移動時間が実移動時間よりも大きくなる軌跡連結候補を仮説生成に用いる軌跡連結候補群から除外するようにした請求項1記載の移動体軌跡識別システム。 - 過去一定時間に検出された移動体の軌跡の組み合わせである軌跡連結候補を生成し、
過去一定時間に検出された移動体の識別情報を軌跡連結候補と組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、その軌跡連結候補・識別情報ペアの集合で所定条件を満たすものを仮説として生成し、
個々の仮説を選択し、選択した仮説に属する軌跡連結候補・識別情報ペアにおいて、その軌跡連結候補が表す軌跡について識別情報が検出される尤もらしさを示す識別情報尤度を計算し、軌跡連結候補・識別情報ペア毎に識別情報尤度を統合することにより、選択した仮説に関する識別情報尤度を算出し、各仮説の識別情報尤度に基づいて最尤仮説を推定する移動体軌跡識別方法。 - 追跡領域内の位置座標に応じた移動体の識別情報の検出確率を予め定義した確率マップを記憶し、
選択された仮説に属する軌跡連結候補・識別情報ペアにおける識別情報の検出時刻毎に、移動体の位置を軌跡連結候補が示す軌跡上に特定し、その位置で移動体が検出される確率値を確率マップから読み取り、その確率値に基づいて識別情報尤度を計算し、更に、軌跡連結候補・識別情報ペア毎に識別情報尤度を統合することにより、選択した仮説に関する識別情報尤度を算出し、
仮説の識別情報尤度に基づいて最尤仮説を推定する請求項12記載の移動体軌跡識別方法。 - 移動体の軌跡と識別情報とが対応付けられる尤もらしさである軌跡・識別情報尤度を記憶し、
最尤仮説とともに生成された全ての仮説に属する軌跡連結候補・識別情報ペアにおいて、その軌跡連結候補に含まれる最新の軌跡と識別情報との組に応じた軌跡・識別情報対応尤度を導出し、
軌跡連結候補・識別情報ペアにおいて、軌跡連結候補に含まれる最新の軌跡と識別情報との組に応じた軌跡・識別情報対応尤度と、識別情報の検出時刻毎に確率マップから読み出した確率値に基づいて、軌跡連結候補・識別情報ペアに関する識別情報尤度を算出するようにした請求項13記載の移動体軌跡識別方法。 - 最尤仮説とともに仮説生成部に生成された全仮説を記憶し、
過去一定時間に検出された識別情報に変化があるか否かを判定し、
識別情報に変化ありと判定された場合に軌跡連結候補を生成して仮説群を生成し、
識別情報に変化ありと判定された場合に生成された仮説群から個々の仮説を選択し、一方、識別情報に変化なしと判定された場合には記憶された仮説群から個々の仮説を選択し
識別情報に変化ありと判定された場合に生成された仮説群から最尤仮説を推定したとき、その最尤仮説とともに生成された全仮説を記憶するようにした請求項13記載の移動体軌跡識別方法。 - 過去一定時間に検出された移動体の属性情報を軌跡連結候補に対応付けるとともに、その属性情報が対応付けられた軌跡連結候補に移動体の識別情報を組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、所定条件を満たす軌跡連結候補・識別情報ペアの集合を仮説として生成し、
選択された仮説に属する軌跡連結候補・識別情報ペアにおいて軌跡連結候補に対応付けられている属性情報が同一の移動体を表す尤もらしさである属性情報尤度を計算し、軌跡連結候補・識別情報ペア毎の属性情報尤度を統合することにより、選択した仮説に関する属性情報を算出し、
仮説の識別情報尤度及び属性情報尤度に基づいて最尤仮説を推定するようにした請求項13記載の移動体軌跡識別方法。 - 過去一定時間に検出された移動体の移動情報を軌跡連結候補に対応付けるとともに、その移動情報が対応付けられた軌跡連結候補に移動体の識別情報を組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、所定条件を満たす軌跡連結候補・識別情報ペアの集合を仮説として生成し、
選択された仮説に属する軌跡連結候補・識別情報ペアにおいて軌跡連結候補に対応付けられている移動情報が同一の移動体を表す尤もらしさである移動情報尤度を計算し、軌跡連結候補・識別情報ペア毎の移動情報尤度を統合することにより、選択した仮説に関する移動情報尤度を算出し、
仮説の識別情報尤度及び移動情報尤度に基づいて最尤仮説を推定するようにした請求項13記載の移動体軌跡識別方法。 - 選択された仮説に属する軌跡連結候補・識別情報ペアにおける軌跡連結候補に含まれている軌跡が連結する尤もらしさである軌跡連結尤度を計算し、
仮説の識別情報尤度及び軌跡連結尤度に基づいて最尤仮説を推定するようにした請求項13記載の移動体軌跡識別方法。 - 軌跡を規定する移動体の位置座標とその検出時刻に基づいて軌跡連結尤度を計算するようにした請求項18記載の移動体軌跡識別方法。
- 軌跡連結候補に含まれている軌跡が連結する尤もらしさである軌跡連結尤度を計算し、その軌跡連結尤度が閾値以上となっている軌跡連結候補を選別し、
選別された軌跡連結候補を用いて仮説群を生成するようにした請求項12記載の移動体軌跡識別方法。 - 追跡領域内に存在する障害物の場所と、その障害物を移動体が通過する際の移動時間とを環境情報として記憶し、
軌跡連結候補における軌跡の消失検出時刻から次の軌跡の出現検出時刻までの実移動時間を計算するとともに、軌跡の消失から次の軌跡の出現までの間に移動体が移動した時間を環境情報に基づいて推定し、その推定移動時間が実移動時間よりも大きくなる軌跡連結候補を仮説生成に用いる軌跡連結候補群から除外するようにした請求項12記載の移動体軌跡識別方法。 - 追跡領域内で移動体が通過可能な経路に沿う移動時間を環境情報として記憶し、
軌跡連結候補における軌跡の消失検出時刻から次の軌跡の出現検出時刻までの実移動時間を計算するとともに、軌跡の消失から次の軌跡の検出までの間に移動体が移動した時間を環境情報に基づいて推定し、その推定移動時間が実移動時間よりも大きくなる軌跡連結候補を仮説生成に用いる軌跡連結候補群から除外するようにした請求項12記載の移動体軌跡識別方法。 - コンピュータに読み込まれて実行される移動体軌跡識別プログラムであって、
過去一定時間に検出された移動体の軌跡の組み合わせである軌跡連結候補を生成し、
過去一定時間に検出された移動体の識別情報を軌跡連結候補と組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、その軌跡連結候補・識別情報ペアの集合で所定条件を満たすものを仮説として生成し、
個々の仮説を選択し、選択した仮説に属する軌跡連結候補・識別情報ペアにおいて、その軌跡連結候補が表す軌跡について識別情報が検出される尤もらしさを示す識別情報尤度を計算し、軌跡連結候補・識別情報ペア毎に識別情報尤度を統合することにより、選択した仮説に関する識別情報尤度を算出し、各仮説の識別情報尤度に基づいて最尤仮説を推定するようにした移動体軌跡識別プログラム。 - 追跡領域内の位置座標に応じた移動体の識別情報の検出確率を予め定義した確率マップを記憶し、
選択された仮説に属する軌跡連結候補・識別情報ペアにおける識別情報の検出時刻毎に、移動体の位置を軌跡連結候補が示す軌跡上に特定し、その位置で移動体が検出される確率値を確率マップから読み取り、その確率値に基づいて識別情報尤度を計算し、更に、軌跡連結候補・識別情報ペア毎に識別情報尤度を統合することにより、選択した仮説に関する識別情報尤度を算出し、
仮説の識別情報尤度に基づいて最尤仮説を推定する請求項23記載の移動体軌跡識別プログラム。 - 移動体の軌跡と識別情報とが対応付けられる尤もらしさである軌跡・識別情報尤度を記憶し、
最尤仮説とともに生成された全ての仮説に属する軌跡連結候補・識別情報ペアにおいて、その軌跡連結候補に含まれる最新の軌跡と識別情報との組に応じた軌跡・識別情報対応尤度を導出し、
軌跡連結候補・識別情報ペアにおいて、軌跡連結候補に含まれる最新の軌跡と識別情報との組に応じた軌跡・識別情報対応尤度と、識別情報の検出時刻毎に確率マップから読み出した確率値に基づいて、軌跡連結候補・識別情報ペアに関する識別情報尤度を算出するようにした請求項24記載の移動体軌跡識別プログラム。 - 最尤仮説とともに仮説生成部に生成された全仮説を記憶し、
過去一定時間に検出された識別情報に変化があるか否かを判定し、
識別情報に変化ありと判定された場合に軌跡連結候補を生成して仮説群を生成し、
識別情報に変化ありと判定された場合に生成された仮説群から個々の仮説を選択し、一方、識別情報に変化なしと判定された場合には記憶された仮説群から個々の仮説を選択し
識別情報に変化ありと判定された場合に生成された仮説群から最尤仮説を推定したとき、その最尤仮説とともに生成された全仮説を記憶するようにした請求項24記載の移動体軌跡識別プログラム。 - 過去一定時間に検出された移動体の属性情報を軌跡連結候補に対応付けるとともに、その属性情報が対応付けられた軌跡連結候補に移動体の識別情報を組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、所定条件を満たす軌跡連結候補・識別情報ペアの集合を仮説として生成し、
選択された仮説に属する軌跡連結候補・識別情報ペアにおいて軌跡連結候補に対応付けられている属性情報が同一の移動体を表す尤もらしさである属性情報尤度を計算し、軌跡連結候補・識別情報ペア毎の属性情報尤度を統合することにより、選択した仮説に関する属性情報を算出し、
仮説の識別情報尤度及び属性情報尤度に基づいて最尤仮説を推定するようにした請求項24記載の移動体軌跡識別プログラム。 - 過去一定時間に検出された移動体の移動情報を軌跡連結候補に対応付けるとともに、その移動情報が対応付けられた軌跡連結候補に移動体の識別情報を組み合わせて軌跡連結候補・識別情報ペアの集合を生成し、所定条件を満たす軌跡連結候補・識別情報ペアの集合を仮説として生成し、
選択された仮説に属する軌跡連結候補・識別情報ペアにおいて軌跡連結候補に対応付けられている移動情報が同一の移動体を表す尤もらしさである移動情報尤度を計算し、軌跡連結候補・識別情報ペア毎の移動情報尤度を統合することにより、選択した仮説に関する移動情報尤度を算出し、
仮説の識別情報尤度及び移動情報尤度に基づいて最尤仮説を推定するようにした請求項24記載の移動体軌跡識別プログラム。 - 選択された仮説に属する軌跡連結候補・識別情報ペアにおける軌跡連結候補に含まれている軌跡が連結する尤もらしさである軌跡連結尤度を計算し、
仮説の識別情報尤度及び軌跡連結尤度に基づいて最尤仮説を推定するようにした請求項24記載の移動体軌跡識別プログラム。 - 軌跡を規定する移動体の位置座標とその検出時刻に基づいて軌跡連結尤度を計算するようにした請求項29記載の移動体軌跡識別プログラム。
- 軌跡連結候補に含まれている軌跡が連結する尤もらしさである軌跡連結尤度を計算し、その軌跡連結尤度が閾値以上となっている軌跡連結候補を選別し、
選別された軌跡連結候補を用いて仮説群を生成するようにした請求項23記載の移動体軌跡識別プログラム。 - 追跡領域内に存在する障害物の場所と、その障害物を移動体が通過する際の移動時間とを環境情報として記憶し、
軌跡連結候補における軌跡の消失検出時刻から次の軌跡の出現検出時刻までの実移動時間を計算するとともに、軌跡の消失から次の軌跡の出現までの間に移動体が移動した時間を環境情報に基づいて推定し、その推定移動時間が実移動時間よりも大きくなる軌跡連結候補を仮説生成に用いる軌跡連結候補群から除外するようにした請求項23記載の移動体軌跡識別プログラム。 - 追跡領域内で移動体が通過可能な経路に沿う移動時間を環境情報として記憶し、
軌跡連結候補における軌跡の消失検出時刻から次の軌跡の出現検出時刻までの実移動時間を計算するとともに、軌跡の消失から次の軌跡の検出までの間に移動体が移動した時間を環境情報に基づいて推定し、その推定移動時間が実移動時間よりも大きくなる軌跡連結候補を仮説生成に用いる軌跡連結候補群から除外するようにした請求項23記載の移動体軌跡識別プログラム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/391,199 US8989438B2 (en) | 2009-08-20 | 2010-08-13 | Mobile body track identification system |
JP2011527661A JP5682563B2 (ja) | 2009-08-20 | 2010-08-13 | 移動体軌跡識別システム、移動体軌跡識別方法および移動体軌跡識別プログラム |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2009191358 | 2009-08-20 | ||
JP2009-191358 | 2009-08-20 | ||
JP2009280926 | 2009-12-10 | ||
JP2009-280926 | 2009-12-10 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2011021588A1 true WO2011021588A1 (ja) | 2011-02-24 |
Family
ID=43607043
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2010/063770 WO2011021588A1 (ja) | 2009-08-20 | 2010-08-13 | 移動体軌跡識別システム |
Country Status (3)
Country | Link |
---|---|
US (1) | US8989438B2 (ja) |
JP (1) | JP5682563B2 (ja) |
WO (1) | WO2011021588A1 (ja) |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2012132285A1 (ja) * | 2011-03-25 | 2012-10-04 | 日本電気株式会社 | 移動体属性推定システム、移動体属性推定方法およびプログラム |
JP2013072858A (ja) * | 2011-09-29 | 2013-04-22 | Panasonic Corp | 移動体位置推定装置、移動体位置推定方法、及び、移動体位置推定プログラム |
JP2013073360A (ja) * | 2011-09-27 | 2013-04-22 | Denso Corp | 隊列走行装置 |
WO2013128852A1 (ja) * | 2012-02-29 | 2013-09-06 | 日本電気株式会社 | 動線情報生成システム、動線情報生成方法および動線情報生成プログラム |
JP2013192154A (ja) * | 2012-03-15 | 2013-09-26 | Omron Corp | 監視装置、信頼度算出プログラム、および信頼度算出方法 |
JP2014048711A (ja) * | 2012-08-29 | 2014-03-17 | Toshiba Corp | 地上移動体経路導出装置 |
CN104036139A (zh) * | 2014-06-12 | 2014-09-10 | 中国科学院软件研究所 | 一种移动对象轨迹监测方法 |
JP2015070354A (ja) * | 2013-09-27 | 2015-04-13 | パナソニックIpマネジメント株式会社 | 移動体追跡装置、移動体追跡システムおよび移動体追跡方法 |
JP2017040530A (ja) * | 2015-08-19 | 2017-02-23 | 株式会社日立製作所 | 移動体計測システム |
JPWO2016139906A1 (ja) * | 2015-03-04 | 2017-12-07 | パナソニックIpマネジメント株式会社 | 人物追跡方法および人物追跡装置 |
WO2018092881A1 (ja) * | 2016-11-21 | 2018-05-24 | 日本電気株式会社 | 動線合成装置、動線合成方法および記録媒体 |
JP6347533B1 (ja) * | 2018-02-16 | 2018-06-27 | タメコ株式会社 | 位置特定方法、位置特定装置およびプログラム |
JP2019166996A (ja) * | 2018-03-23 | 2019-10-03 | 株式会社Ihi | 監視装置及び監視方法 |
JP2020057424A (ja) * | 2019-12-20 | 2020-04-09 | 株式会社東芝 | 移動物体追跡装置、方法及びプログラム |
JP2020144140A (ja) * | 2020-05-19 | 2020-09-10 | 株式会社 ミックウェア | サーバ、及び情報処理方法 |
WO2020261378A1 (ja) * | 2019-06-25 | 2020-12-30 | 日本電気株式会社 | 軌跡連結装置、軌跡連結方法、及び、プログラムが格納された非一時的なコンピュータ可読媒体 |
JPWO2021245749A1 (ja) * | 2020-06-01 | 2021-12-09 | ||
US20220196431A1 (en) * | 2020-12-18 | 2022-06-23 | Denso Corporation | Driving assistance system, driving assistance apparatus, driving assistance method, image recognition apparatus, and image recognition method |
US11395096B2 (en) | 2018-06-04 | 2022-07-19 | Nec Corporation | Information sharing device, information sharing method, and recording medium |
US11614339B2 (en) | 2017-04-13 | 2023-03-28 | Micware Co., Ltd. | Information processing device and recording medium |
KR102576587B1 (ko) * | 2022-09-16 | 2023-09-11 | 주식회사 노타 | 다중 카메라를 이용한 이동체 궤적 생성 방법 및 장치 |
Families Citing this family (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8989438B2 (en) * | 2009-08-20 | 2015-03-24 | Nec Corporation | Mobile body track identification system |
JP5919665B2 (ja) * | 2011-07-19 | 2016-05-18 | 日本電気株式会社 | 情報処理装置、物体追跡方法および情報処理プログラム |
CN103703497B (zh) * | 2011-07-26 | 2016-08-31 | 丰田自动车株式会社 | 车辆确定系统及车辆确定装置 |
US9292936B2 (en) * | 2013-01-09 | 2016-03-22 | Omiimii Ltd. | Method and apparatus for determining location |
JP6433149B2 (ja) * | 2013-07-30 | 2018-12-05 | キヤノン株式会社 | 姿勢推定装置、姿勢推定方法およびプログラム |
US10152864B2 (en) * | 2014-02-28 | 2018-12-11 | Tyco Fire & Security Gmbh | Distributed rules engines for robust sensor networks |
US10026015B2 (en) * | 2014-04-01 | 2018-07-17 | Case Western Reserve University | Imaging control to facilitate tracking objects and/or perform real-time intervention |
WO2016151950A1 (ja) | 2015-03-20 | 2016-09-29 | 日本電気株式会社 | 監視システム、監視方法および監視プログラム |
JP6678512B2 (ja) * | 2016-05-17 | 2020-04-08 | 株式会社日立製作所 | センサ設置位置支援装置 |
US10018476B2 (en) | 2016-08-17 | 2018-07-10 | Apple Inc. | Live traffic routing |
US10060753B2 (en) * | 2016-08-17 | 2018-08-28 | Apple Inc. | On-demand shortcut computation for routing |
EP3493102B1 (en) * | 2017-11-30 | 2020-04-29 | Axis AB | A method and system for tracking a plurality of objects in a sequence of images |
WO2019180917A1 (ja) * | 2018-03-23 | 2019-09-26 | 日本電気株式会社 | オブジェクト追跡装置、オブジェクト追跡方法、及びオブジェクト追跡プログラム |
JP2020102135A (ja) * | 2018-12-25 | 2020-07-02 | 清水建設株式会社 | 追跡システム、追跡処理方法、及びプログラム |
US11555910B2 (en) * | 2019-08-02 | 2023-01-17 | Motional Ad Llc | Merge-split techniques for sensor data filtering |
KR102307217B1 (ko) * | 2019-12-12 | 2021-09-30 | (주)위즈베이스 | 드론 촬영 영상을 이용하여 도로 교통 정보를 분석하는 방법 및 이를 이용한 장치 |
JP7405657B2 (ja) * | 2020-03-17 | 2023-12-26 | 本田技研工業株式会社 | 移動体監視システム、及び移動体監視方法 |
CN113706574B (zh) * | 2020-05-20 | 2024-01-02 | 杭州海康威视数字技术股份有限公司 | 一种移动轨迹确定方法、装置、电子设备及存储介质 |
US20210391908A1 (en) * | 2020-06-10 | 2021-12-16 | Peking University | Method, system, computer device, and storage medium for non-contact determination of a sensing boundary |
EP3937065B1 (en) * | 2020-07-07 | 2022-05-11 | Axis AB | Method and device for counting a number of moving objects that cross at least one predefined curve in a scene |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005250692A (ja) * | 2004-03-02 | 2005-09-15 | Softopia Japan Foundation | 物体の同定方法、移動体同定方法、物体同定プログラム、移動体同定プログラム、物体同定プログラム記録媒体、移動体同定プログラム記録媒体 |
JP2007087150A (ja) * | 2005-09-22 | 2007-04-05 | Matsushita Electric Ind Co Ltd | 画像再生方法、画像再生機、画像再生プログラム |
JP2008122093A (ja) * | 2006-11-08 | 2008-05-29 | Mitsubishi Electric Corp | 多目標追尾装置 |
JP2008535041A (ja) * | 2005-02-04 | 2008-08-28 | ブリティッシュ・テレコミュニケーションズ・パブリック・リミテッド・カンパニー | 映像フレーム内におけるスプリアス領域の識別 |
Family Cites Families (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003139856A (ja) | 2001-10-31 | 2003-05-14 | Hitachi Ltd | 目標統合検出装置 |
US7058205B2 (en) * | 2001-12-07 | 2006-06-06 | Xerox Corporation | Robust, on-line, view-based appearance models for visual motion analysis and visual tracking |
US6954544B2 (en) * | 2002-05-23 | 2005-10-11 | Xerox Corporation | Visual motion analysis method for detecting arbitrary numbers of moving objects in image sequences |
JP4450306B2 (ja) | 2003-07-11 | 2010-04-14 | Kddi株式会社 | 移動体追跡システム |
JP2006352626A (ja) | 2005-06-17 | 2006-12-28 | Nippon Telegr & Teleph Corp <Ntt> | 移動軌跡分析装置およびプログラム |
EP1990788B1 (en) * | 2006-03-01 | 2012-09-05 | Toyota Jidosha Kabushiki Kaisha | Vehicle path determining method and vehicle course determining device |
JP5147036B2 (ja) * | 2006-03-08 | 2013-02-20 | 株式会社国際電気通信基礎技術研究所 | 位置推定装置、位置推定方法及び位置推定プログラム |
EP1879149B1 (en) * | 2006-07-10 | 2016-03-16 | Fondazione Bruno Kessler | method and apparatus for tracking a number of objects or object parts in image sequences |
JP2008299742A (ja) | 2007-06-01 | 2008-12-11 | Mitsubishi Electric Corp | 人物追跡装置 |
JP4951490B2 (ja) | 2007-12-18 | 2012-06-13 | 日本電信電話株式会社 | 動物体追跡装置、動物体追跡方法及び動物体追跡プログラム並びに動物体追跡プログラムを記録した記録媒体 |
US8374388B2 (en) * | 2007-12-28 | 2013-02-12 | Rustam Stolkin | Real-time tracking of non-rigid objects in image sequences for which the background may be changing |
US8989438B2 (en) * | 2009-08-20 | 2015-03-24 | Nec Corporation | Mobile body track identification system |
US8970701B2 (en) * | 2011-10-21 | 2015-03-03 | Mesa Engineering, Inc. | System and method for predicting vehicle location |
-
2010
- 2010-08-13 US US13/391,199 patent/US8989438B2/en active Active
- 2010-08-13 JP JP2011527661A patent/JP5682563B2/ja active Active
- 2010-08-13 WO PCT/JP2010/063770 patent/WO2011021588A1/ja active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005250692A (ja) * | 2004-03-02 | 2005-09-15 | Softopia Japan Foundation | 物体の同定方法、移動体同定方法、物体同定プログラム、移動体同定プログラム、物体同定プログラム記録媒体、移動体同定プログラム記録媒体 |
JP2008535041A (ja) * | 2005-02-04 | 2008-08-28 | ブリティッシュ・テレコミュニケーションズ・パブリック・リミテッド・カンパニー | 映像フレーム内におけるスプリアス領域の識別 |
JP2007087150A (ja) * | 2005-09-22 | 2007-04-05 | Matsushita Electric Ind Co Ltd | 画像再生方法、画像再生機、画像再生プログラム |
JP2008122093A (ja) * | 2006-11-08 | 2008-05-29 | Mitsubishi Electric Corp | 多目標追尾装置 |
Cited By (35)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPWO2012132285A1 (ja) * | 2011-03-25 | 2014-07-24 | 日本電気株式会社 | 移動体属性推定システム、移動体属性推定方法およびプログラム |
WO2012132285A1 (ja) * | 2011-03-25 | 2012-10-04 | 日本電気株式会社 | 移動体属性推定システム、移動体属性推定方法およびプログラム |
JP5907162B2 (ja) * | 2011-03-25 | 2016-04-20 | 日本電気株式会社 | 移動体属性推定システム、移動体属性推定方法およびプログラム |
JP2013073360A (ja) * | 2011-09-27 | 2013-04-22 | Denso Corp | 隊列走行装置 |
JP2013072858A (ja) * | 2011-09-29 | 2013-04-22 | Panasonic Corp | 移動体位置推定装置、移動体位置推定方法、及び、移動体位置推定プログラム |
WO2013128852A1 (ja) * | 2012-02-29 | 2013-09-06 | 日本電気株式会社 | 動線情報生成システム、動線情報生成方法および動線情報生成プログラム |
JPWO2013128852A1 (ja) * | 2012-02-29 | 2015-07-30 | 日本電気株式会社 | 動線情報生成システム、動線情報生成方法および動線情報生成プログラム |
US10648803B2 (en) | 2012-02-29 | 2020-05-12 | Nec Corporation | Movement line information generation system, movement line information generation method and movement line information generation program |
US10895454B2 (en) | 2012-02-29 | 2021-01-19 | Nec Corporation | Movement line information generation system, movement line information generation method and movement line information generation program |
JP2013192154A (ja) * | 2012-03-15 | 2013-09-26 | Omron Corp | 監視装置、信頼度算出プログラム、および信頼度算出方法 |
JP2014048711A (ja) * | 2012-08-29 | 2014-03-17 | Toshiba Corp | 地上移動体経路導出装置 |
JP2015070354A (ja) * | 2013-09-27 | 2015-04-13 | パナソニックIpマネジメント株式会社 | 移動体追跡装置、移動体追跡システムおよび移動体追跡方法 |
CN104036139A (zh) * | 2014-06-12 | 2014-09-10 | 中国科学院软件研究所 | 一种移动对象轨迹监测方法 |
JPWO2016139906A1 (ja) * | 2015-03-04 | 2017-12-07 | パナソニックIpマネジメント株式会社 | 人物追跡方法および人物追跡装置 |
JP2017040530A (ja) * | 2015-08-19 | 2017-02-23 | 株式会社日立製作所 | 移動体計測システム |
US11023739B2 (en) | 2016-11-21 | 2021-06-01 | Nec Corporation | Flow line combining device, flow line combining method, and recording medium |
JPWO2018092881A1 (ja) * | 2016-11-21 | 2019-08-08 | 日本電気株式会社 | 動線合成装置、動線合成方法およびプログラム |
WO2018092881A1 (ja) * | 2016-11-21 | 2018-05-24 | 日本電気株式会社 | 動線合成装置、動線合成方法および記録媒体 |
US11614339B2 (en) | 2017-04-13 | 2023-03-28 | Micware Co., Ltd. | Information processing device and recording medium |
JP2019144025A (ja) * | 2018-02-16 | 2019-08-29 | タメコ株式会社 | 位置特定方法、位置特定装置およびプログラム |
JP6347533B1 (ja) * | 2018-02-16 | 2018-06-27 | タメコ株式会社 | 位置特定方法、位置特定装置およびプログラム |
JP2019166996A (ja) * | 2018-03-23 | 2019-10-03 | 株式会社Ihi | 監視装置及び監視方法 |
JP7212455B2 (ja) | 2018-03-23 | 2023-01-25 | 株式会社Ihi | 監視装置及び監視方法 |
US11395096B2 (en) | 2018-06-04 | 2022-07-19 | Nec Corporation | Information sharing device, information sharing method, and recording medium |
WO2020261378A1 (ja) * | 2019-06-25 | 2020-12-30 | 日本電気株式会社 | 軌跡連結装置、軌跡連結方法、及び、プログラムが格納された非一時的なコンピュータ可読媒体 |
JPWO2020261378A1 (ja) * | 2019-06-25 | 2020-12-30 | ||
JP7164040B2 (ja) | 2019-06-25 | 2022-11-01 | 日本電気株式会社 | 軌跡連結装置、軌跡連結方法、及び、プログラム |
JP2020057424A (ja) * | 2019-12-20 | 2020-04-09 | 株式会社東芝 | 移動物体追跡装置、方法及びプログラム |
JP2020144140A (ja) * | 2020-05-19 | 2020-09-10 | 株式会社 ミックウェア | サーバ、及び情報処理方法 |
JP7050114B2 (ja) | 2020-05-19 | 2022-04-07 | 株式会社 ミックウェア | サーバ、及び情報処理方法 |
WO2021245749A1 (ja) * | 2020-06-01 | 2021-12-09 | 日本電気株式会社 | 追跡装置、追跡方法、および記録媒体 |
JPWO2021245749A1 (ja) * | 2020-06-01 | 2021-12-09 | ||
JP7359306B2 (ja) | 2020-06-01 | 2023-10-11 | 日本電気株式会社 | 追跡装置、追跡システム、追跡方法、およびプログラム |
US20220196431A1 (en) * | 2020-12-18 | 2022-06-23 | Denso Corporation | Driving assistance system, driving assistance apparatus, driving assistance method, image recognition apparatus, and image recognition method |
KR102576587B1 (ko) * | 2022-09-16 | 2023-09-11 | 주식회사 노타 | 다중 카메라를 이용한 이동체 궤적 생성 방법 및 장치 |
Also Published As
Publication number | Publication date |
---|---|
JP5682563B2 (ja) | 2015-03-11 |
JPWO2011021588A1 (ja) | 2013-01-24 |
US20120148102A1 (en) | 2012-06-14 |
US8989438B2 (en) | 2015-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5682563B2 (ja) | 移動体軌跡識別システム、移動体軌跡識別方法および移動体軌跡識別プログラム | |
JP6070689B2 (ja) | 動線情報生成システム、動線情報生成方法および動線情報生成プログラム | |
JP5807635B2 (ja) | 動線検出システム、動線検出方法および動線検出プログラム | |
CN110782483B (zh) | 基于分布式相机网络的多视图多目标跟踪方法及系统 | |
WO2011102416A1 (ja) | 移動物体追跡システムおよび移動物体追跡方法 | |
Papaioannou et al. | Fusion of radio and camera sensor data for accurate indoor positioning | |
Xia et al. | Robust unsupervised factory activity recognition with body-worn accelerometer using temporal structure of multiple sensor data motifs | |
Kaâniche et al. | Recognizing gestures by learning local motion signatures of HOG descriptors | |
JP7339314B2 (ja) | 室内測位および追跡のための深層スマートフォンセンサ融合 | |
Li et al. | Robust multiperson detection and tracking for mobile service and social robots | |
Cao et al. | Correlation-based tracking of multiple targets with hierarchical layered structure | |
CN102200578A (zh) | 数据关联设备和方法 | |
US10614436B1 (en) | Association of mobile device to retail transaction | |
WO2015049340A1 (en) | Marker based activity transition models | |
Ogris et al. | Continuous activity recognition in a maintenance scenario: combining motion sensors and ultrasonic hands tracking | |
CN112381853A (zh) | 用于利用无线信号和图像来进行人员检测、跟踪和标识的装置和方法 | |
Tan et al. | Implicit multimodal crowdsourcing for joint RF and geomagnetic fingerprinting | |
CN113988228A (zh) | 一种基于rfid和视觉融合的室内监控方法及系统 | |
JP2019194758A (ja) | 情報処理装置、情報処理方法、およびプログラム | |
CN107665495B (zh) | 对象跟踪方法及对象跟踪装置 | |
Wietrzykowski et al. | Adopting the FAB-MAP algorithm for indoor localization with WiFi fingerprints | |
Kim et al. | InFo: indoor localization using fusion of visual information from static and dynamic cameras | |
KR101340287B1 (ko) | 스마트 홈에서 마이닝 기반의 패턴 분석을 이용한 침입탐지 시스템 | |
Wu et al. | Indoor surveillance video based feature recognition for pedestrian dead reckoning | |
CN115331120A (zh) | 一种无人机目标人员跟踪方法及系统 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 10809933 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13391199 Country of ref document: US Ref document number: 2011527661 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 10809933 Country of ref document: EP Kind code of ref document: A1 |