JP6336935B2 - Moving object tracking device - Google Patents

Description

  The present invention relates to a moving object tracking device that tracks a moving object in a time-series image.

  For the purpose of security or the like, a person is tracked with time-series images periodically taken by a surveillance camera. In that case, occlusion where a part of the person is hidden, sudden movement of the person's movement, etc. are factors that reduce the continuity of the tracking.

  Therefore, in the moving object tracking device described in Patent Document 1 below, a plurality of templates having different parts (parts) in the person image are generated for each person, and each template is tracked. In this way, even if some parts are lost due to occlusion, the remaining parts can be used to track a moving object.

  Moreover, in the moving object tracking device described in Patent Document 2 below, each person is tracked using a particle filter. According to the particle filter, a large number of hypotheses representing the predicted position of the person are diffused around the position where the person is predicted to be moved, and the predicted position where the characteristic of the person appears most in the actual captured image Is determined as the position of the person. By spreading this hypothesis, high-precision tracking can be maintained even if there is a sudden change in the movement of a person.

JP 2008-250772 A JP 2010-49296 A

  From the techniques of Patent Documents 1 and 2, by setting multiple parts to a person and tracking each part simultaneously with a particle filter, a person who has undergone a large posture change accompanying part occlusion or sudden movement change like a fall It is possible to prevent a decrease in tracking continuity.

  However, if a plurality of parts are set for each person, the number of objects to be processed increases, so the balance between calculation cost and tracking accuracy becomes a problem. That is, in order to track a person in real time, it is necessary to complete the processing for all parts of particles within the shooting period, so the total number of particles that can be set is limited. And, if the number of particles that can be set is limited, if you keep setting a certain number of particles in parts that are difficult to track due to occlusion etc., the number of particles that are effective for tracking will decrease, and the tracking accuracy of the whole person will be improved There was a problem to lower.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a moving object tracking device that can accurately track a moving object with a large posture change without increasing the calculation cost.

  The moving object tracking device according to the present invention is for tracking a moving object in a time-series image, and for each of a plurality of parts constituting the moving object, the image feature, the past position, and the likelihood of the past position are displayed. Storage means for storing past tracking reliability, candidate position calculating means for calculating one or more candidate positions at the current time from the past position for each part, and current time for each candidate position of the part A likelihood calculating means for calculating the likelihood of the candidate position by comparing the image feature in the image and the image feature stored in the storage means, and weighting using the likelihood for each part Position calculating means for calculating the position of the moving object at the current time using the candidate position, the candidate position calculating means at the current time of each part of the moving object. The number of the candidate positions, determined in accordance with the past tracking reliability of the parts.

  In the moving object tracking device according to another aspect of the present invention, the candidate position calculation means determines in advance the total number of candidate positions to be calculated for each moving object, and depends on the past tracking reliability of each part. The total number is distributed to each part.

  In another moving object tracking device according to the present invention, for each of the parts, a variation degree of the candidate position obtained by weighting each candidate position using the likelihood and a peak of the spatial distribution of the likelihood And a tracking reliability calculation means for determining the past tracking reliability according to the degree of the sharpness distribution, and further determining the likelihood distribution between the plurality of parts. The distribution ratio of the number of candidate positions in is determined in accordance with the ratio of the past tracking reliability.

  Further, the moving object tracking device according to another aspect of the present invention deletes a lost part whose past tracking reliability is lower than a predetermined lower limit value among the plurality of parts registered in the storage unit, and the position calculating unit The system further comprises a part setting means for defining a new part different from the lost part corresponding to the calculated position of the moving object and registering the new part in the storage means, and the candidate position calculation means for the new part, A predetermined initial number of candidate positions are set.

  According to the present invention, since a limited number of particles can be allocated to a part that is effective for tracking among a plurality of parts of a moving object, the moving object can be tracked with high accuracy without increasing the calculation cost. .

1 is a block diagram showing a schematic configuration of a moving object tracking device according to an embodiment of the present invention. It is a block diagram explaining the function of the memory | storage part and tracking part in the moving object tracking apparatus which concerns on embodiment of this invention. It is a schematic diagram which shows the example of arrangement | positioning of an imaging part. It is a schematic diagram explaining tracking person information. Is a schematic diagram showing an example of tracking parts at some time during the tracking is set to the first person H _1. It is a schematic diagram which shows the example of the tracking part information corresponding to the tracking part shown in FIG. It is a schematic diagram which shows the example of the particle information in a certain time during tracking. It is explanatory drawing which shows the example of distribution of a particle. It is a schematic diagram which shows the example of the relationship between likelihood distribution and tracking reliability. FIG. 5 is an overall flowchart showing an outline of the operation of the moving object tracking device according to the embodiment of the present invention. It is a general | schematic flowchart of the tracking process in the moving object tracking device which concerns on embodiment of this invention. It is a general | schematic flowchart of the new tracking person setting process in the moving object tracking device which concerns on embodiment of this invention.

  Hereinafter, as an example of a preferred embodiment of the present invention, a moving object tracking device 1 that tracks a person as a moving object using a fixed camera will be described.

[Configuration of Moving Object Tracking Device 1]
FIG. 1 is a block diagram showing a schematic configuration of the moving object tracking device 1. The moving object tracking device 1 includes an imaging unit 2, a storage unit 3, a tracking unit 4, and an output unit 5. FIG. 2 is a block diagram illustrating the functions of the storage unit 3 and the tracking unit 4.

The imaging unit 2 is a so-called monitoring camera, and includes an imaging element such as a CCD element or a C-MOS element, an optical system component, an A / D converter, and the like. FIG. 3 is a schematic diagram illustrating an arrangement example of the imaging unit 2. For example, the imaging unit 2 is installed so as to face the persons H ₁ and H ₂ in the monitoring space from above. The imaging unit 2 is connected to the tracking unit 4, images the monitoring space at predetermined time intervals, and sequentially outputs the captured time-series monitoring images to the tracking unit 4. Hereinafter, the unit of time recorded at the predetermined time interval is referred to as time.

  The storage unit 3 is a storage unit composed of a memory device such as a ROM (Read Only Memory) or a RAM (Random Access Memory). The storage unit 3 stores various programs and various data, and inputs and outputs such information to and from the tracking unit 4. To do. Various data stored in the storage unit 3 includes tracking person information 20, tracking part information 21, and particle information 22.

FIG. 4 is a schematic diagram for explaining the tracking person information 20. FIG. 4A is an example of a monitoring image at a certain time during tracking, and two persons H ₁ and H ₂ are shown as tracking targets. Incidentally, as shown in the figure, the horizontal direction and the vertical direction of the monitoring image are the x-axis and y-axis, respectively, and the position in the monitoring image is represented by a set (x, y) of x-coordinate and y-coordinate. FIG. 4B shows the tracking person information 20 corresponding to the monitoring image of FIG. 4A. The tracking person information 20 is the identification number (ID) of the person to be tracked and the estimation of the person in the monitoring image. The table 200 is stored in association with the position 200. Specifically, the tracking person IDs of the persons H ₁ and H ₂ are 1 and 2, respectively, and the estimated position can be defined, for example, at the position of the center of gravity of the person image or the top of the head.

Figure 5 is a schematic view showing an example of tracking parts are set on the person H ₁ at a certain time during the tracking, Figure 6 is a schematic illustrating an example of a tracking part information 21 that corresponds to the track parts shown in FIG. 5 FIG. For example, the tracking part R _j (j = 1 to 10) is set for the person H ₁ with the tracking part ID as j, and a part thereof is shown in FIG. The tracking part information 21 is associated with the image feature, the past position, the past tracking reliability indicating the likelihood of the past position, and the like for each tracking part, and is stored in a table format, for example. In FIG. 6, the template image is an example of the image feature of the tracking part, and instead of or in addition to the template image, the tracking histogram information such as a color histogram calculated from the template image or a feature amount such as a luminance gradient is used as the image feature. 21 can be stored. The template image of the tracking part is an image obtained by cutting out an area corresponding to the tracking part from the monitoring image. The “position” shown in FIG. 6 is a part tracking position (past position) at a past time, for example, one time before the current time, and is represented by coordinates (x, y) in the monitoring image. The “relative position with the tracking person” is the coordinates of the reference point of the tracking part with respect to the reference point of the person (for example, the center of gravity of the person image), and the “tracking reliability” is the past tracking reliability.

  FIG. 7 is a schematic diagram showing an example of the particle information 22 at a certain time during tracking. The particle information 22 correlates the tracking person ID, particle ID, tracking part ID, particle position (x, y), velocity (vx, vy), likelihood, and the like, and is stored in, for example, a table format.

  The tracking unit 4 is an arithmetic device such as a CPU (Central Processing Unit), a DSP (Digital Signal Processor), or an MCU (Micro Control Unit). The tracking unit 4 reads out a program from the storage unit 3 and executes the program, whereby a tracking person setting unit 40, a tracking part setting unit 41, a candidate position calculation unit 42, a position calculation unit 43, a likelihood calculation unit 44, which will be described later, It functions as a tracking reliability calculation unit 45, a tracking end unit 46, etc., processes a monitoring image input from the imaging unit 2 to detect a person in the monitoring space, tracks the detected person with a particle filter, and tracks The result is output to the output unit 5.

  The output unit 5 is a display device such as a display device, and displays the person tracking result output from the tracking unit 4.

  The tracking person setting unit 40 detects a person who has not yet started tracking in the monitoring image, and newly stores the detected position as an estimated position of the tracking person information 20 in addition to the tracking target. The person detection may be performed by using a known method such as a method of labeling a change area of an image or a method of performing machine learning using luminance gradient information, and thus detailed description thereof is omitted.

  The tracking part setting unit 41 (part setting unit) determines parts to be tracked by the particle filter for the person set as the tracking target by the tracking person setting unit 40, and generates tracking part information 21 for each part. Store in the storage unit 3. Parts can be selected by using a predetermined method such as the head, left shoulder, right shoulder, left hip, right hip, and abdomen. The method of determining can be used. The tracking reliability is not determined immediately after setting the tracking parts, but is calculated by the tracking reliability calculation means 45 after the start of tracking and stored in the storage unit 3 (step S23 in FIG. 11 described later).

  Further, the tracking part setting unit 41 determines a lost part that is difficult to continue tracking among the tracking parts based on the tracking reliability stored in the tracking part information 21, and the tracking part deleting unit excludes the lost part from the tracking target. It has a function. For example, the determination can be performed by threshold processing for the tracking reliability, and the tracking part setting unit 41 deletes a plurality of parts whose tracking reliability is lower than a predetermined lower limit value as a lost part. Specifically, the tracking part setting unit 41 determines the particle group related to the lost part from the tracking part ID of each particle in the particle information 22 and deletes the particle group from the particle information 22. Further, the tracking part setting unit 41 deletes information on the lost part from the tracking part information 21.

  Further, the tracking part setting unit 41 has a function of a tracking part adding unit that adds a tracking part to a tracking person according to a situation. As described above, the parts determined to be difficult to continue tracking are deleted. As a result, the number of tracking parts decreases as the tracking of the person continues, and this decrease in the number of tracking parts can be a factor that reduces the robustness against posture fluctuation and concealment. Therefore, the tracking part setting means 41 adds tracking parts when the number of tracking parts falls below a certain number so that the number of tracking parts does not decrease extremely. The tracking parts to be added are preferably selected so as to be arranged evenly with respect to the person in consideration of the parts currently being tracked. Particle groups are generated corresponding to the added tracking parts. The particle information 22 for the particle group is initially registered based on the tracking part information 21 and the like. The particle speed can be set to the movement speed of the person position.

  The candidate position calculation unit 42 performs particle generation processing for generating a fixed number of particles (candidate positions) for the new tracking person set by the tracking person setting unit 40. In this embodiment, the number of particles per tracking person is 500. The candidate position calculation means 42 assigns the tracking parts set by the tracking part setting means 41 to each particle and initializes the state variables of each particle. The particle state variable includes position, velocity, likelihood, and tracking part ID. In this initialization, the ID and position of the assigned tracking part are set in the particle information 22, and 0 is set because the speed is still unknown. In addition, the likelihood is not obtained immediately after the particle generation, but is calculated by the likelihood calculating means 44 described later and stored in the particle information 22. Allocating tracking parts to particles in a new tracking person is a random method, or a method of setting a predetermined ratio for each part such as 30% head, 20% abdomen, 20% left shoulder, and 20% right shoulder. It is also good. Here, since the number of particles is fixed for the tracking person, the calculation cost is constant regardless of the number of tracking parts.

  In addition, the candidate position calculation unit 42 calculates one or more particles at the current time from the past position for each tracking part. That is, the candidate position calculation unit 42 functions as a sampling unit, performs motion prediction on the input image using the past position of the tracking part for each particle, calculates the position at the current time, and uses the predicted result. The position / velocity of the state variable is updated and stored in the particle information 22. For example, a method of adding random noise after performing estimation using a motion model such as constant velocity linear motion can be used for motion prediction.

  Further, the candidate position calculation unit 42 has a function as the distribution adjustment unit 420. The distribution adjusting means 420 determines the number of particles at the current time of each part of the tracking person according to the tracking reliability of the part. For example, the distribution adjusting unit 420 determines in advance the total number of particles to be calculated for each tracking person, and distributes the total number to each part according to the tracking reliability of each part. Specifically, more particles are assigned to parts with higher tracking reliability. For example, the distribution ratio of the number of particles among a plurality of parts can be set according to the ratio of tracking reliability. In this embodiment, 500 particles per person to be tracked are tracked reliability between tracking parts. Are distributed so that the ratio of particles and the ratio of the number of particles are equal. FIG. 8 is an explanatory diagram showing an example of particle distribution. As illustrated in FIG. 8, the distribution is performed such that the ratio of the tracking reliability of each part is equal to the ratio of the number of particles after the distribution adjustment.

  When tracking parts are added by the tracking part setting means 41, the number of particles corresponding to the value obtained by dividing the total number of particles by the total number of tracking parts is distributed to the newly added parts, and the remaining particles are allocated. Distribution can be made according to the ratio of the tracking reliability described above. For example, when the number of parts is 10, 10% of the total number of particles is assigned to the new part. As another example, a predetermined fixed number of particles may be assigned to a new part.

  Further, the candidate position calculation unit 42 performs resampling of particles for each tracking part based on the likelihood of each particle obtained by the likelihood calculation unit 44 and the number of particles for each tracking part allocated by the distribution adjustment unit 420. It functions as a sampling means. Resampling is a process aimed at removing many particles with low likelihood and duplicating particles with high likelihood to leave many particles near the position where the tracking target is likely to exist. is there. Candidate position calculation means 42 is realized by selecting and duplicating from the current particle group based on the likelihood for the number of particles allocated to each tracking part. For selection of particles based on likelihood, a selection operation method in a genetic algorithm can be used. For example, a method such as roulette selection can be used. The candidate position calculation means 42 updates the particle information 22 with the resampled particle group after the resampling is completed.

  The position calculation means 43 estimates the position of the tracking person at the current time. The position calculation means 43 first estimates the position of the tracking person for each tracking part. The person position estimation for each tracking part is performed using particles related to the tracking part. For example, for each tracking part, the position calculation means 43 calculates the person position for each particle using the relative position information between the tracking part and the tracking person stored in the tracking part information 21, and the particle position is determined for the person position. The estimated position of the person for each tracking part is calculated by performing a weighted average with the likelihood of.

  Next, the position calculation means 43 integrates the estimated positions for each tracking part to obtain the final estimated position of the tracking person. For example, the position calculation means 43 performs a weighted average with the tracking reliability of each tracking part as a weight for the estimated position for each tracking part, and calculates the final estimated position of the tracking person. The position calculation means 43 stores the estimated position of the tracked person finally obtained in the tracked person information 20.

  The likelihood calculating means 44 calculates the likelihood of each particle. The likelihood calculating means 44 calculates the feature amount of the input image at the current time at the position of the particle, and compares the tracking part assigned to the particle with the feature amount of the template stored in the tracking part information 21. Calculate the degree. Specifically, a color histogram, a luminance gradient similarity, or the like can be used as the feature amount. The calculated likelihood is stored in the particle information 22.

  The tracking reliability calculation means 45 calculates the reliability (tracking reliability) of each tracking part. For each tracking part, the tracking reliability calculation means 45 estimates the likelihood based on the degree of variation in position in the particle group obtained by weighting each particle using the likelihood and the peak height of the spatial distribution of likelihood. The degree of sharpness of the degree distribution is evaluated, and the past tracking reliability is determined according to the degree of sharpness. That is, the tracking reliability is calculated based on the likelihood distribution approximated by the position and likelihood of the particle group of the tracking part. If the spatial distribution of likelihood is unimodal and high, it can be considered that there is a high possibility that tracking is performed correctly. Therefore, it is desirable to define the tracking reliability so as to be high when the likelihood distribution is a distribution having a single peak and a high peak. Whether or not the likelihood distribution is unimodal can be evaluated based on the degree of variation of the particle group in which the likelihood value is added as a weight. The height of the likelihood distribution can be evaluated from the likelihood value of the particles near the distribution peak.

  FIG. 9 is a schematic diagram showing an example of the relationship between the likelihood distribution and the tracking reliability. The position of the particle group is actually defined by two-dimensional coordinates corresponding to the monitoring image, but in FIG. 9, it is simplified to one-dimensional coordinates with the horizontal direction as the coordinate axis. The vertical direction in the figure corresponds to the likelihood value, and the likelihood value increases as it goes upward. A likelihood distribution 450a shown in FIG. 9A is an example of a distribution having a single peak and a high peak value. In such a case, the tracking reliability is set high. On the other hand, the likelihood distribution 450b shown in FIG. 9B is an example of a distribution that is unimodal but has a low peak value, and the likelihood distribution 450c shown in FIG. 9C is an example of a multimodal distribution. Yes, the tracking reliability is set low in such cases.

  Specifically, as a method of determining the tracking reliability, for example, (Method 1) a method using the total likelihood value of particles around the estimated position of the tracking part, or (Method 2) weighted variance and the likelihood of all particles There is a method using the total value of.

(Method 1) First, an estimated position of the tracking part is calculated by a weighted average using the likelihood as a weight, and the total value of the likelihood of the particle group included in a certain range around the estimated position (for example, within a circle of a predetermined radius) omega and _C, the ratio omega _{C /} omega relative likelihood total value of all the particles of tracking parts omega and evaluation values for unimodal. When the distribution is high in unimodality, the likelihood distribution has a peak at the above-described estimated position, and the likelihood value in the vicinity thereof becomes high. Therefore, the value of the ratio of the likelihood total value ω _C / ω increases. Further, the total value of likelihood within a certain range around the estimated position is an evaluation value indicating the height of the peak of the likelihood distribution. Finally, the tracking reliability is the product of the evaluation value for the single peak and the evaluation value indicating the peak height. The tracking reliability is an evaluation value indicating that the likelihood distribution is unimodal and the peak of the likelihood distribution is high. The size of the surrounding area is determined based on the size of the tracking part, and may be determined to be 30% of the radius of the tracking part.

  (Method 2) The evaluation value for unimodality is set as an evaluation value based on a weighted variance value with respect to the position of the particle having the likelihood as a weight, and the evaluation value indicating the peak height is set as the total value of the likelihood of all particles. The tracking reliability is the product of the evaluation value for the single peak and the evaluation value indicating the height of the peak. If the variance value is low and the total likelihood value of all particles is high, it can be considered that particles with high likelihood are concentrated in one place, that is, the peak height is high with unimodality. it can.

  Alternatively, as a modification of the above-described method, only particles having a likelihood value equal to or greater than a certain value are selected (that is, after threshold cutting), and the distribution of the positions of the selected particles and the total number of particles The tracking reliability may be defined using When the variance of particles with a certain likelihood or higher is low and the total number is large, it can be considered that particles with high likelihood are gathered in one place, and the likelihood distribution is unimodal and peaks. Can be considered high.

  Alternatively, when the calculation cost is reduced or the processing is simplified, the total likelihood value of all the particle groups may be simply set as the tracking reliability. However, in this case, since it is not evaluated to be unimodal, the accuracy can be reduced accordingly.

  The tracking reliability calculation means 45 stores the tracking reliability calculated as described above in the tracking part information 21.

  The tracking end means 46 determines a situation in which further tracking is impossible, such as when a person is no longer reflected on the camera due to movement outside the screen or concealment of structures, or when tracking fails. The tracking process for the person is terminated. The determination that tracking is impossible uses a combination of a method of determining when there is no tracking part having a tracking reliability of a certain value or more and a method of determining when the main position of the particle is outside the screen. If it is determined that tracking is not possible, information regarding the person is deleted from the tracking person information 20, the tracking part information 21, and the particle information 22.

[Operation of Moving Object Tracking Device 1]
Hereinafter, the operation of the moving object tracking device 1 will be described. FIG. 10 is a schematic overall flowchart of the operation of the moving object tracking device 1. When the operation of the moving object tracking device 1 is started, the imaging unit 2 acquires a monitoring image at a predetermined frame rate (for example, 7 fps) and outputs the monitoring image to the tracking unit 4. The tracking unit 4 repeats a series of processes in steps S1 to S5 each time a monitoring image is input (step S1).

  The tracking unit 4 performs a tracking process on the input monitoring image for each person recorded in the tracking person information 20, and estimates the current position (step S2). The tracking unit 4 sequentially selects the tracking target person recorded in the tracking person information 20 as the target of the tracking process S2, and when the tracking process S2 is completed for all the tracking target persons, the tracking unit 4 performs the process in step S4. On the other hand, if there is an unprocessed person to be tracked, the tracking process S2 is continued (step S3).

  FIG. 11 is a schematic flowchart of the tracking process S2, and the process will be described in more detail with reference to FIG.

  The tracking unit 4 functions as a sampling unit by the candidate position calculation unit 42 and samples particles for each tracking person (step S21). The tracking unit 4 calculates the likelihood of each particle by the likelihood calculating means 44 based on the sampling result in step S21 (step S22). Further, the tracking unit 4 calculates the tracking reliability of each tracking part by the tracking reliability calculation means 45 based on the likelihood of the particle group obtained in step S22 (step S23).

  The tracking end means 46 determines whether or not the tracking can be continued based on the tracking reliability of the parts and the position of the particle group calculated in step S23 (step S24). This is performed (step S25). As a result, the tracking of the person who is determined not to be tracked is terminated, and the tracking end unit 46 deletes the information related to the person from the tracked person information 20.

  On the other hand, with respect to the person determined to be traceable in step S24, first, if there is a tracking part that is difficult to keep track of, it is deleted (step S26). That is, the tracking unit 4 functions as a tracking part deleting unit by the tracking part setting unit 41, and determines a lost part that is difficult to continue tracking among the tracking parts based on the tracking reliability stored in the tracking part information 21, Exclude lost parts from tracking.

  Thereafter, the tracking unit 4 functions as the position calculating unit 43 and estimates the position of the tracking person based on the position of the particle group sampled in Step S21 and the tracking reliability of each tracking part calculated in Step S23 (Step S21). S27).

  The tracking unit 4 functions as a tracking part adding unit by the tracking part setting unit 41, and adds a tracking part according to the status of the tracking part (step S28).

  Then, the tracking unit 4 determines the distribution of the number of particles for each tracking part based on the tracking reliability of each tracking part calculated in step S23 by the distribution adjusting unit 420 (step S29). Based on the distributed number of particles, the tracking unit 4 resamples particles for each tracking part by the candidate position calculation means 42 (step S30). However, resampling is not necessary for particles relating to the tracking part newly added in step S28 because it is the first time.

  When the tracking process S2 described above is performed for all the persons registered in the tracked person information 20, the tracking unit 4 proceeds to step S4 in FIG. A person who has not been set for tracking in the monitoring image is detected, and if it is detected, it is added as a tracking person (step S4).

  FIG. 12 is a schematic flowchart of the new tracking person setting process S4. This process will be described in more detail with reference to FIG.

  The tracking person setting unit 40 detects a person who has not yet been registered in the tracking person information 20 as a tracking target in the monitoring image, and adds the person to the tracking person information 20 (step S41). The tracking person setting means 40 performs tracking part setting (step S42) and particle group generation (step S43) for each newly detected person, and tracking part setting processing S42 and particle generation for all new persons. When the process S43 is completed, the tracking unit 4 ends the new tracked person setting process and advances the process to step S5 in FIG. 10 (step S44). In step S42, the tracking part setting function of the tracking part setting unit 41 sets a part to be tracked for a person to be newly tracked. In step S43, the candidate position calculation unit 42 generates particles. The function generates a particle group for tracking by the particle filter using the tracking part information set in step S42 for the person who is newly tracked.

  When the tracking of the person is completed by the above-described processes S2 to S4 for the monitoring image input in step S1, the tracking unit 4 outputs the tracking result to the output unit 5 (step S5). For example, the tracking unit 4 displays the estimated positions of all persons on the display unit of the output unit 5 as a tracking result.

  As described above, in the present invention, when tracking a moving object by setting a plurality of parts on a moving object and tracking each part using a particle filter, the total number per moving object is determined in advance. A particle group is distributed to each tracking part according to its likelihood. In other words, in order to realize real-time tracking, it is necessary to complete the tracking process for the monitoring image during a relatively short frame period, so the number of particles that can be set per frame is limited. . Therefore, in the present invention, instead of assigning the same number of particles to the parts set as moving objects, the number of assignments to difficult-to-follow parts is reduced, and that part is assigned to parts that are likely to continue tracking. Here, if any part can be tracked, the tracking accuracy of the moving object can be maintained. Therefore, according to the present invention, it is possible to track a moving object with high accuracy without increasing the amount of calculation.

  DESCRIPTION OF SYMBOLS 1 Moving object tracking device, 2 Imaging part, 3 Storage part, 4 Tracking part, 5 Output part, 20 Tracking person information, 21 Tracking part information, 22 Particle information, 40 Tracking person setting means, 41 Tracking part setting means, 42 Candidate Position calculation means, 43 Position calculation means, 44 Likelihood calculation means, 45 Tracking reliability calculation means, 46 Tracking end means, 420 Distribution adjustment means

Claims (4)

A moving object tracking device for tracking a moving object in a time-series image,
Storage means for storing the image feature, the past position, and the past tracking reliability representing the likelihood of the past position for each of a plurality of parts constituting the moving object;
Candidate position calculating means for calculating one or more candidate positions at the current time from the past position for each part;
For each candidate position of the part, likelihood calculating means for calculating the likelihood of the candidate position by comparing the image feature in the image at the current time with the image feature stored in the storage means;
Position calculating means for calculating the position of the moving object at the current time using the candidate position weighted using the likelihood for each part;
Have
The candidate position calculating means determines the number of candidate positions at the current time of each part of the moving object according to the past tracking reliability of the part;
A moving object tracking device. The moving object tracking device according to claim 1,
The candidate position calculating means is configured to determine in advance a total number of the candidate positions to be calculated for each moving object, and to distribute the total number to each part according to the past tracking reliability of each part; Feature moving object tracking device. In the moving object tracking device according to claim 1 or 2,
The degree of sharpness of the likelihood distribution using the degree of variation of the candidate position obtained by weighting each candidate position using the likelihood and the height of the peak of the spatial distribution of the likelihood for each part. Further comprising tracking reliability calculation means for determining the past tracking reliability according to the degree of sharpness,
The candidate position calculating means determines a distribution ratio of the number of candidate positions among the plurality of parts according to a ratio of the past tracking reliability;
A moving object tracking device. In the moving object tracking device according to any one of claims 1 to 3,
Of the plurality of parts registered in the storage unit, the lost part whose past tracking reliability is lower than a predetermined lower limit value is deleted, and the part corresponding to the position of the moving object calculated by the position calculation unit is deleted. It further has a part setting means for defining a new part different from the lost part and registering it in the storage means,
The candidate position calculating means sets a predetermined initial number of candidate positions for the new part;
A moving object tracking device.

Images