JP2013072858A - Mobile object position estimation device, mobile object position estimation method and mobile object position estimation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a mobile object position estimation device and method and a mobile object position estimation program, which do not cause an estimation error of a mobile route of an object without causing lowering of position estimation accuracy even in a scene in which a correspondence between an observed value and the object becomes ambiguous due to passing of mobile objects with each other, or the like.SOLUTION: Identification information allocation means 103 allocates identification information included in observed values of first observation means 101 to observed values of second observation means 102 so that closeness in position between the observed values of the first observation means 101 and the observed values of the second observation means 102 becomes minimum at a point A of time when the observed values can be obtained from the first observation means 101 for outputting the identification information and positional information of the mobile object with a low frequency and at a point B of time when the observed values can be obtained from the observation means 101 at the next time of the point A of time and the total of movement amount of the observed values of the observation means 102 becomes minimum in a time section AB with respect to a series of observed values obtained from the second observation means 102 for outputting the positional information of the mobile object with a high frequency at the point A of time and the point B of time and in the time section AB.

Description

  The present invention relates to a mobile object position estimation device and a mobile object position estimation device that estimate positions of a plurality of moving objects (moving objects such as people) existing in a space to be observed based on observation values relating to a target object from a sensor. The present invention relates to a method and a moving object position estimation program.

  As a method for estimating the position of a moving body such as a person based on sensor observation values, there is a method that complements the weaknesses of sensing with only a single sensor by integrating the observation values of different types of sensors with different sensing performance. To do.

  For example, when trying to identify a person and estimate the position using a camera and an image sensor based on image recognition processing, it is often difficult to identify a plurality of people with high accuracy from each other only with the image recognition processing. Therefore, apart from the camera, combining the observation value of the sensor (wireless tag) with high human identification performance and the observation value of the camera eliminates the weak point of the single sensor and identifies the moving object. There is a conventional technique (Non-patent Document 1) that performs estimation.

  In Non-Patent Document 1, although the position information of the object can be output, the observation value of the sensor by the camera without image recognition capability and the sensor by image recognition, and the observation of the sensor by the wireless tag which has high object identification capability and can also output the position information A method is disclosed in which a moving locus of an object is output by estimating a position that changes sequentially while integrating moving values while identifying moving objects.

  Specifically, the method disclosed in Non-Patent Document 1 includes, in an online Bayesian estimation framework, an association variable that probabilistically represents a correspondence between a plurality of observation values and a plurality of position estimation distributions in the estimation process. Is used to simultaneously estimate the position of a moving body such as a plurality of people at the same time.

  More specifically, the distance between the ID likelihood, which is a discrete distribution of identification information (ID) of the object, and the observed position of the object and the estimated position distribution of each object within the estimation process. The value obtained by normalizing the product of two values of distance likelihood calculated from (the distance likelihood takes a larger value because the closer the distance is, the more likely the correspondence between the observed value and the position estimation distribution is) Calculated as an association value.

  This association value is one value assigned to the correspondence between the observed value and the estimated position distribution, and how much the position information included in each observed value should contribute to the update of the estimated position of each object. It corresponds to the weight of.

  The position estimation is performed in consideration of the ambiguity of the correspondence between the observed value and the object by updating the estimated distribution with a value obtained by weighting the observed value position of each observed value with the association value.

  In particular, Non-Patent Document 1 is for estimating the movement trajectory of a plurality of objects based on the association value even when a sensor that cannot output identification information (ID) (for example, a camera sensor having a recognition function) is used. The configuration and method are disclosed. It is difficult to identify the object with high accuracy in the image recognition processing of the imaging data of the camera, but whether or not two observation values that are temporally different in one observation period observed the same object It can be easily discriminated by, for example, evaluating the similarity of the image features of the object imaged at the time.

  Therefore, if it can be determined that the same object is observed before and after the observation period, the ID likelihood that the camera sensor cannot output the correspondence between the observed value and the object at the previous observation time is maintained. Is approximated by an association value obtained based on previous observations. On the other hand, if it is determined that the objects are not the same before and after the observation period, the correspondence between the observed value and the object at the previous observation time is not maintained, and therefore the ID likelihood that the camera sensor cannot output is Approximate with a uniform distribution (the ID likelihood being a uniform distribution corresponds to the fact that the correspondence between the observed value and the object is unknown). When the determination state of whether the objects are the same is ambiguous, the determination state is expressed by a probability value (tracking likelihood), and the two cases are combined by the tracking likelihood, and the ID likelihood is calculated. Approximate.

(Equation 1)
ID likelihood of camera sensor observation value at that time = tracking likelihood × previous association value + (1−tracking likelihood) × uniform distribution
. . . . (Formula 1)

  As described above, by approximating the ID likelihood shown in Equation 1, the position of the plurality of objects can be estimated by giving the approximate value of the identification information to the sensor that cannot output the identification information of the target object. .

Toru Tanikawa, Katsuyoshi Yamagami, “Development of mobile tracking technology by integrating sensor information of camera and wireless tag”, 28th Annual Conference of the Robotics Society of Japan, September 22-24, 2010

  However, in the method disclosed in Non-Patent Document 1, in a situation where the distance between moving objects that are observation targets is close, the correspondence between the observation value and the target object becomes ambiguous, so that the position estimation is performed. There is a problem that errors accumulate and the moving path of the moving object is erroneously estimated.

  This problem will be described using an example in which two people move so as to cross each other.

  FIG. 17 shows a person who moves from north to south in the space 1701 (a person whose ID is “A” in FIG. 17) and a person who moves from the west to the east (a person whose ID is “B” in FIG. 17). It is a figure which shows a person's movement path | route 1702, 1703 at the time of moving so that it may pass in the middle vicinity.

  1701A in FIG. 17 is an area near the center of the space 1701 represented by a broken-line rectangle. A phenomenon occurring in the region 1701A near the center will be described with reference to FIG.

  18A, 18B, and 18C show the position estimation distribution (indicated by a circle in the figure) of the person A and the person B in the area corresponding to the area 1701A near the center in FIG. Observed values (indicated by triangles in the figure). Assume that the observation proceeds in the order of (A), (B), and (C).

  In the method disclosed in Non-Patent Document 1, in a situation where moving objects are close to each other, a position estimation distribution in which the observation value truly corresponds in terms of distance likelihood based on the proximity between the observation value and the position estimation distribution. And the distance likelihood value of the position estimation distribution whose observation value does not truly correspond to the distance likelihood value. For example, in the situation shown in FIG. 18B, the distance between the camera observation value of person A and the position estimation distribution of person A, the camera observation value of person A, and the position estimation distribution of person B. The distance likelihood between the A person's camera observation value and the A person's position estimation distribution, the A person's camera observation value, and the B person's position estimation The distance likelihood value with the distribution is relatively close.

  Further, in the method disclosed in Non-Patent Document 1, it is ambiguous whether or not the same object can be observed because it becomes difficult to separate the two moving objects in the image identification process in a situation where the moving objects approach each other. Become. Therefore, the probability value of the tracking likelihood of Equation 1 becomes a small value, and as a result, the ratio that the uniform distribution is reflected in the ID likelihood becomes large, and the observed value is true from the viewpoint of the approximated ID likelihood. The value of the ID likelihood with respect to the observed object and the value of the ID likelihood with respect to other than the object whose observed value is truly observed are relatively close to each other.

  When the distance likelihood value and the ID likelihood value approach evenly, as a result, the association value calculated from the product of the distance likelihood and the ID likelihood also corresponds to the position estimated distribution. And the association value between the observed value that should not correspond to the position estimation distribution are relatively close to each other.

  For this reason, when the position estimation distribution is updated, the ratio of the position information of observations other than the observations that actually observed the target object contributes to the update relatively. The position estimation distribution is updated in a direction deviating from the direction, and the position estimation error increases.

  The solid line arrow in FIG. 18 illustrates the direction in which the position estimation distribution is updated with the observed value. In the situation of FIG. 18 (B), the position estimation distribution of the person A, the position estimation distribution of the person B, and the updated directions are the position of the camera observation value of the person A and the camera observation value of the person B, respectively. It is shifted from the direction of the position. For this reason, at the next observation time (situation (C)), the position estimation distribution of the person A shifts to the east, and the position estimation distribution of the person B shifts to the south.

  In the situation of FIG. 18C, the distance between the observed value and the position estimated distribution that should correspond to the observed value is separated, rather, the distance between the observed value and the position estimated distribution that should not correspond to the observed value is close. As a result, the direction of the position estimation distribution based on the observed values is further incorrect, and the position estimation error further increases.

  As described above, in the method of Non-Patent Document 1, a position estimation error is likely to occur particularly in a situation where an object is approaching, and a path completely different from a path on which a person has actually moved is estimated due to the error. It has the problem of end up.

  FIG. 19 shows the movement paths 1702 and 1703 of the two people A and B as a result of position estimation based on the method of Non-Patent Document 1 with respect to the movement of the two people A and B shown in FIG. It is a figure which shows the result of having estimated the completely different movement path | route accidentally.

  As shown in FIG. 18, due to an increase in position estimation error in the vicinity where two people A and B approach and intersect, the movement history of the position estimation distribution of each person A and B does not intersect, This is the result of moving in the wrong direction near the center.

  When further proceeding from the situation of (C) in FIG. 18, with respect to the RFID tag observation value that outputs the identification information of the object, the position information output by the observation value and the object corresponding to the identification information of the observation value The distance from the position estimation distribution is also far away, and conversely, a situation occurs in which the position estimation distribution of an object that does not correspond to the identification information of the observed value is close.

  In the method disclosed in Non-Patent Document 1, the weight likelihood (association value) for updating the estimated value of the position information of the object with the position information of the observed value is calculated using the distance likelihood (the position information of the observed value and the target). Since the calculation is based on the product of the proximity of the object position estimation distribution) and the ID likelihood, the weight in the ID likelihood is discounted by the distance likelihood. Therefore, the identification information of the RFID tag observation value, which has high object identification performance and should contribute to the correction of the position estimation error, does not contribute effectively to the correction of the estimation error of the position estimation distribution. Therefore, once the position estimation distribution greatly deviates from the position of the true person, the deviation increases, resulting in a problem that leads to completely different estimation results.

  Furthermore, when the observation frequency of the wireless tag is lower than the observation frequency of the camera, the contribution of the update to the estimated position of the target object of the observation value of the wireless tag from the viewpoint of time is further weakened. There is a problem that is difficult to occur.

  In addition, Non-Patent Document 1 discloses only a method in which camera observation values are always obtained by the number of people, and a case where camera observation values cannot be obtained for the number of people, that is, disappearance of camera observation values or cameras There is no description of how to handle cases where more observations are obtained than the actual number of people. In the human recognition processing by the camera, the observation value of the camera may not be obtained for the number of people due to the reason that the person to be observed is blocked by the shielding object or that two persons approaching each other are detected as one person. Further, it may happen that a camera observation value is obtained more than the number of people by misrecognizing an object that is not a person as a person. However, Non-Patent Document 1 does not disclose a method for dealing with such a case, and if the camera observation values disappear and the number of observation values is less than the number of people, one camera observation value is converted to the observation value. This is reflected in the update of the position estimation distribution of a plurality of people including those who do not correspond to the above, and has the problem of causing an erroneous estimation result.

  Therefore, an object of the present invention is to solve the above-described problem, without causing a decrease in position estimation accuracy even in a scene in which the correspondence between the observed value and the target object is ambiguous, such as a moving object passing through, and Another object of the present invention is to provide a mobile object position estimation device, a mobile object position estimation method, and a mobile object position estimation program that do not cause an estimation error of a moving path of an object.

  In order to solve the above-described problems, the present invention is configured as follows.

According to one aspect of the present invention, a plurality of moving objects existing in an observation space are observed, and observation values including the identification information of the moving object and the position information of the moving object that are sequentially obtained for each moving object are input. As the identification information of the plurality of mobile objects that are observation targets, and a mobile object position estimation device that estimates position information,
First observation means for outputting identification information and position information of the mobile body as observation values at predetermined time intervals;
Second observation means for outputting position information of the moving body as an observation value at a time interval different from the interval of the first observation means;
Position information of the observed value of the first observing means and the observation of the second observing means in an observation section of the time interval of each of the first observing means and the second observing means. The total of the relative distance of the position information of the value and the total amount of movement of the observation value of the second observation unit in the observation section are minimized with respect to the observation value of the second observation unit. Identification information assignment means for assigning identification information obtained from the first observation means;
Position estimation means for estimating the position of the moving object based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means The
A mobile object position estimation apparatus having the same is provided.

According to another aspect of the present invention, a plurality of moving objects existing in an observation space are observed, and an observation value including the identification information of the moving object and the position information of the moving object that are sequentially obtained for each moving object is input. As an identification information of the plurality of mobile objects that are observation targets, and a mobile object position estimation method for estimating position information,
The identification information and position information of the mobile object are output as observation values by the first observation means at predetermined time intervals,
The position information of the moving body is output as an observation value by the second observation unit at a time interval different from the interval of the first observation unit,
Position information of the observed value of the first observing means and the observation of the second observing means in an observation section of the time interval of each of the first observing means and the second observing means. The total of the relative distance of the position information of the value and the total amount of movement of the observation value of the second observation unit in the observation section are minimized with respect to the observation value of the second observation unit. The identification information obtained from the first observation means is assigned by the identification information assignment means,
Based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means, the position of the moving body is estimated by the position estimation means To
A mobile object position estimation method is provided.

According to yet another aspect of the present invention, a plurality of mobile objects existing in an observation space are observed, and an observation value including the identification information of the mobile object and the position information of the mobile object obtained sequentially for each mobile object is obtained. As an input, a plurality of mobile object identification information to be observed, and a mobile object position estimation program for estimating position information,
First observation means for outputting the identification information and position information of the mobile object as observation values at predetermined time intervals, and a time different from the interval of the first observation means using the position information of the mobile object as observation values Position information of the observed value of the first observing means and position of the observed value of the second observing means in a certain observation section of each time interval with the second observing means outputting at intervals The first relative to the observation value of the second observation means is such that the total of the relative distance of information and the total amount of movement of the observation value of the second observation means in the observation section are minimized. Identification information assigning means for assigning identification information obtained from the observation means;
Position estimation means for estimating the position of the moving object based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means A moving object position estimation program for causing a computer to realize the function is provided.

  The mobile object position estimation apparatus, the mobile object position estimation method, and the mobile object position estimation program of the present invention do not have an observation object identification capability based on identification information of a sensor having a high observation object identification capability ( Appropriate identification information can be assigned to the observation value of the sensor (or the discrimination ability is low), so the correspondence between the observation value and the observation object can always be determined. Therefore, even in scenes where the correspondence between the observed value and the target object is ambiguous, such as when the moving object passes, there is no particular effect that the position estimation accuracy will not decrease and that the target movement path will not be erroneously estimated. Have

  In addition, the mobile object position estimation device, the mobile object position estimation method, and the mobile object position estimation program of the present invention have the ability to identify an observation object even in a situation where an observation value disappears or over-observation occurs due to an observation error. Appropriate identification information can be assigned to observation values of sensors that do not have (or have low identification ability). Therefore, even in a situation where the observation target is not always correctly sensed in the actual environment, there is a special effect that the position estimation accuracy is not lowered and an error in estimating the moving path of the target object is not caused. The

The block diagram which shows the structure of the moving body position estimation apparatus of 1st Embodiment of this invention. The block diagram which shows the structure of the human position tracking apparatus based on the mobile body position estimation apparatus of 1st Embodiment of this invention. The figure which shows the example of the condition which detects the position of a person with the person position tracking apparatus based on 1st Embodiment of this invention. The figure which shows an example of the observed value recorded on the observed value buffer based on 1st Embodiment of this invention. The figure which shows an example of the observed value recorded on the observed value buffer based on 1st Embodiment of this invention. The figure which shows the example of the person position information recorded on the person position estimated value recording part buffer based on 1st Embodiment of this invention. The figure which shows the example of the position of a person displayed on the person position display part, and a movement locus | trajectory based on 1st Embodiment of this invention. The flowchart of the main process of the person position tracking process based on 1st Embodiment of this invention. The figure which shows the example of a movement locus | trajectory of the person for demonstrating the processing method of step S105 based on 1st Embodiment of this invention. The figure which shows the example of the observed value used as the object of ID allocation processing for demonstrating the processing method of step S105 based on 1st Embodiment of this invention. The figure which shows the formulation of ID allocation to the observation value in the certain time t for demonstrating the processing method of step S105 based on 1st Embodiment of this invention. The figure which shows the combination of ID allocation to the observation value in the certain time t for demonstrating the processing method of step S105 based on 1st Embodiment of this invention. Explanatory drawing that ID allocation process of the process of step S105 becomes a minimum cost route search problem based on 1st Embodiment of this invention. The figure explaining the prediction cost setting in the case of solving the minimum cost route search problem of the ID allocation process of the process of step S105 based on 1st Embodiment of this invention with an A * algorithm. The figure which shows the example of the information stored in each node at the time of solving the minimum cost path | route search problem of the ID allocation process of the process of step S105 based on 1st Embodiment of this invention with an A * algorithm. The flowchart of the operation | movement regarding creation of the start node in the case of solving the minimum cost path | route search problem of the ID allocation process of the process of step S105 based on 1st Embodiment of this invention, and creation of a target node. The flowchart of the procedure which derives | leads-out the minimum cost path | route when solving the minimum cost path | route search problem of ID allocation process of the process of step S105 based on 1st Embodiment of this invention with an A * algorithm. The figure explaining the method of pruning the hypothesis node of ID allocation by the minimum cost path | route search problem of ID allocation process of the process of step S105 based on 1st Embodiment of this invention. The figure explaining the method of pruning the hypothesis node of ID allocation by the minimum cost path | route search problem of ID allocation process of the process of step S105 based on 1st Embodiment of this invention. The figure which shows the example of the movement locus | trajectory of two moving bodies for demonstrating the subject of the person position estimation in a prior art. The figure which shows the cause of a human position estimation mistake in the positional relationship of two moving bodies and observation values for demonstrating the subject in a prior art. The figure which shows the example of the estimation error of the person position movement path | route in a prior art. The block diagram which shows the structure of the human position tracking apparatus based on 2nd Embodiment of this invention. The flowchart of the main process of the person position tracking process based on 2nd Embodiment of this invention. The flowchart which shows the flow of the process which produces | generates ID allocation hypothesis in step S105A based on 2nd Embodiment of this invention. The figure which shows the example of the person recognition camera observation value in connection with the processing content of the observation value movement possibility determination part based on 2nd Embodiment of this invention. The figure which shows the example of the determination result of the observation value movement possibility determination part based on 2nd Embodiment of this invention. The figure which shows the example of the processing content of the virtual observation value production | generation part based on 2nd Embodiment of this invention. The figure which shows the example which ID allocation control part based on 2nd Embodiment of this invention allocated ID with respect to the observed value. The figure showing the node expression of ID allocation hypothesis based on 2nd Embodiment of this invention. The figure which shows the example of the estimation result of the person position movement path | route in 1st Embodiment of this invention.

  Embodiments according to the present invention will be described below in detail with reference to the drawings.

  DESCRIPTION OF EMBODIMENTS Hereinafter, various embodiments of the present invention will be described before detailed description of embodiments of the present invention with reference to the drawings.

According to the first aspect of the present invention, a plurality of moving objects existing in an observation space are observed, and an observation value including the identification information of the moving object and the position information of the moving object that are sequentially obtained for each moving object is input. As the identification information of the plurality of mobile objects that are observation targets, and a mobile object position estimation device that estimates position information,
First observation means for outputting identification information and position information of the mobile body as observation values at predetermined time intervals;
Second observation means for outputting position information of the moving body as an observation value at a time interval different from the interval of the first observation means;
Position information of the observed value of the first observing means and the observation of the second observing means in an observation section of the time interval of each of the first observing means and the second observing means. The total of the relative distance of the position information of the value and the total amount of movement of the observation value of the second observation unit in the observation section are minimized with respect to the observation value of the second observation unit. Identification information assignment means for assigning identification information obtained from the first observation means;
Position estimation means for estimating the position of the moving object based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means The
A mobile object position estimation apparatus having the same is provided.

According to the second aspect of the present invention, the first observing means receives the radio wave including the identification information of the moving body and transmitted from the radio tag of the moving body, and receives the identification information and position of the moving body. An RFID tag observation unit that outputs information as an observation value, or an ultrasonic sensor that transmits ultrasonic waves to the moving body and receives reflected ultrasonic waves and outputs identification information and position information of the moving body as observation values And
The second observing means captures an image and outputs position information of the moving body as an observation value, or receives a millimeter wave transmitted from the moving body and reflected from the moving body, The millimeter wave transmitter / receiver that outputs position information as an observation value, or a pressure sensor that is arranged on a floor surface of the observation space and outputs position information of the moving body as an observation value. A mobile object position estimation apparatus is provided.

  According to the third aspect of the present invention, in the identification information assigning means, the observed value of the second observing means that is outside the movable range in which the moving body is movable within the time of the observation interval is deleted, Based on the observation value of the second observation means within the movable range within which the mobile body can move within the observation interval, the previous observation value of the second observation means when performing the identification information assignment process An observation value that cannot be moved from the position of the second observation means when the identification information assignment is performed, or one that can be moved to the position of the observation value of the second observation means when the identification information assignment is performed A first observation value for when the identification information assignment is performed is specified, and the identification information obtained from the first observation means is assigned to the specified observation value of the second observation means. Or 2 aspects Providing mobile location estimation device according.

  With this configuration, observation errors assumed in the second observation means, for example, disappearance of observation values due to concealment of observation objects or proximity of observation objects, or objects other than observation objects are erroneous It is possible to grasp the state of occurrence of erroneous observation values caused by detection, and among the observation values of the second observation means, the identification information obtained from the first observation means should or should not be assigned. It can be distinguished from things. Therefore, it is possible to avoid assignment of erroneous identification information due to an observation error, and it is possible to prevent a decrease in accuracy of the final position estimation result of the moving body due to the observation error.

  According to a fourth aspect of the present invention, the identification information assigning means is identified as having no one that can be moved to the position of the observation value of the second observation means when performing the identification information assignment, Based on the movement history of the observation value with respect to the previous observation value when the identification information assignment is performed, a movement destination position when the identification information assignment is performed is predicted, and a virtual observation value is set at the prediction position. The moving body position estimation apparatus according to the third aspect, which includes a virtual observation value generation unit that generates

  Assuming that the observation object has moved to a position that is considered to be appropriate from the movement history of the observation object, this configuration virtually generates an observation value and assigns identification information obtained from the first observation means. Thus, even when the observation value of the second observation means of the observation target is not obtained at a certain observation time due to the reason that the observation target is concealed, the movement path of the observation target can be estimated. Is possible.

  According to the fifth aspect of the present invention, in the identification information allocating means, the movable range of the moving object to be observed is set to a maximum moving speed based on a measurement result of the moving ability of the moving object, and the observation. The moving body position estimation apparatus according to the third or fourth aspect is set, wherein the range is a circle having a radius of the product of the intervals.

  According to a sixth aspect of the present invention, in the identification information assigning means, the movable range of the mobile object to be observed is assigned to the identification information assignment of the movement history of the observation value of the second observation means. Centered on the destination position obtained on the assumption that the movement speed and the movement direction calculated from the position of the previous observation value with respect to the time of performing and the position of the previous observation value with respect to the time of assigning the identification information are maintained The moving body position estimating apparatus according to the third or fourth aspect is defined as a circle having a predetermined radius.

According to the seventh aspect of the present invention, a plurality of moving objects existing in the observation space are observed, and an observation value including the identification information of the moving object and the position information of the moving object that are sequentially obtained for each moving object is input. As an identification information of the plurality of mobile objects that are observation targets, and a mobile object position estimation method for estimating position information,
The identification information and position information of the mobile object are output as observation values by the first observation means at predetermined time intervals,
The position information of the moving body is output as an observation value by the second observation unit at a time interval different from the interval of the first observation unit,
Position information of the observed value of the first observing means and the observation of the second observing means in an observation section of the time interval of each of the first observing means and the second observing means. The total of the relative distance of the position information of the value and the total amount of movement of the observation value of the second observation unit in the observation section are minimized with respect to the observation value of the second observation unit. The identification information obtained from the first observation means is assigned by the identification information assignment means,
Based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means, the position of the moving body is estimated by the position estimation means To
A mobile object position estimation method is provided.

According to an eighth aspect of the present invention, the first observation value including identification information and position information of the mobile object and output from the first observation means, including the position information of the mobile object and the first From the first time when the second observation value output from the second observation means is obtained simultaneously, the second time when the first observation value and the second observation value are obtained simultaneously. For the second observation obtained during
At the first time, the identification information assigning means obtains a combination having the shortest distance between the position of the first observation value and the position of the second observation value,
Based on the combination at the first time, the identification information of the first observation value is allocated by the identification information allocation means as the identification information of the second observation value,
Further, at the second time, the identification information assigning means obtains a combination having the shortest total distance between the position of the first observation value and the position of the second observation value,
In the combination at the second time, the identification information of the first observation value is assigned by the identification information assigning means as the identification information of the second observation value,
In the identification information allocating means, the allocation state of the identification information to the second observation value at the first time is a start node, and the identification information to the second observation value at the second time is A node that is located between the start node and the target node, with a combination of identification information assigned to the second observation value between the first time and the second time, with the assignment state as the target node In the search graph obtained by setting the transition cost between the nodes as the total movement amount of the second observation value, the minimum cost path from the start node to the target node of the search graph is determined. By obtaining, the identification information allocating means assigns the identification information of the first observation value to the second observation value from the first time to the second time.
A moving body position estimation method according to a seventh aspect is provided.

  According to the ninth aspect of the present invention, in the search processing on the search graph, the identification information assigning means determines whether the assignment state from the node corresponding to the assignment state of the identification information at a certain observation time to the next observation time. When generating a node corresponding to a hypothesis, a correspondence relationship between observation values of the second observation means that can move within the observation interval is determined in consideration of the movable range of the moving object to be observed. The previous observation for the time when the search processing is performed by specifying the observation value when the search processing that is impossible to move from the position of the observation value of the second observation means for the previous time when the search processing is performed The moving body position estimation method according to the eighth aspect, which does not generate an assigned hypothesis node including an unmovable observation value from a value position.

  According to the tenth aspect of the present invention, in the search processing on the search graph, the identification information assigning means determines the assignment state from the node corresponding to the assignment state of the identification information at a certain observation time to the next observation time. When generating a node corresponding to a hypothesis, it is determined that there is no one that can move to the position of the observation value when performing the search process, considering the movable range of the mobile object to be observed, Based on the movement history of the observation value with respect to the previous observation value when the search process is performed, a movement destination position when the search process is performed is predicted, and a virtual observation value is generated at the prediction position And the moving body position estimation method as described in the 8th or 9th aspect which produces | generates the hypothesis node that the said said observed value moved to the said virtual observed value is provided.

According to the eleventh aspect of the present invention, a plurality of moving objects existing in an observation space are observed, and an observation value including the identification information of the moving object and the position information of the moving object that are sequentially obtained for each moving object is input. As an identification information of the plurality of mobile objects that are observation targets, and a mobile object position estimation program that estimates position information,
First observation means for outputting the identification information and position information of the mobile object as observation values at predetermined time intervals, and a time different from the interval of the first observation means using the position information of the mobile object as observation values Position information of the observed value of the first observing means and position of the observed value of the second observing means in a certain observation section of each time interval with the second observing means outputting at intervals The first relative to the observation value of the second observation means is such that the total of the relative distance of information and the total amount of movement of the observation value of the second observation means in the observation section are minimized. Identification information assigning means for assigning identification information obtained from the observation means;
Position estimation means for estimating the position of the moving object based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means A moving object position estimation program for causing a computer to realize the function is provided.

According to a twelfth aspect of the present invention, the first observation value including identification information and position information of the mobile object and output from the first observation means, including the position information of the mobile object and the first From the first time when the second observation value output from the two observation means is obtained at the same time, the second observation value is obtained at the same time as the first observation value and the second observation value are obtained simultaneously. For the second observation obtained during the time,
A function for obtaining a combination of the shortest distance between the position of the first observation value and the position of the second observation value by the identification information assigning means at the first time;
A function of assigning identification information of the first observation value as identification information of the second observation value by the identification information assigning means based on the combination at the first time;
A function for obtaining a combination of the shortest distance between the position of the first observation value and the position of the second observation value by the identification information assigning unit at the second time;
A function of assigning the identification information of the first observation value by the identification information assigning means as the identification information of the second observation value in the combination at the second time;
The assignment state of identification information to the second observation value at the first time is a start node, and the assignment state of identification information to the second observation value at the second time is a target node, Search obtained by setting a combination of assignment of identification information to the second observation value between the first time and the second time as a node located between the start node and the target node In the graph, by setting the transition cost between the nodes as the total amount of movement of the second observation value, the minimum cost path from the start node to the target node of the search graph is obtained, thereby obtaining the first cost. A function of assigning identification information of the first observation value to the second observation value from the time of the second time to the second time by the identification information assigning means;
A moving body position estimation program according to an eleventh aspect for causing a computer to realize the above is provided.

  Embodiments according to the present invention will be described below in detail with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram showing the configuration of the mobile object position estimation apparatus according to the first embodiment of the present invention. The mobile object position estimation apparatus includes first observation means 101, second observation means 102, identification information assignment means 103, and position estimation means 104.

  The first observation unit 101 outputs identification information for distinguishing a plurality of observation targets and position information of the observation targets to the identification information assignment unit 103 and the position estimation unit 104.

  The second observation unit 102 outputs only position information to the identification information allocation unit 103 for the same observation target as the first observation unit 101.

  The identification information assigning means 103 is an observation value that associates the correspondence between each observation value of the first observation means 101 with high object identification accuracy and each observation value of the second observation means 102 that cannot identify the object. Correlate so that the total distance is the smallest. Based on the correspondence in the identification information assigning unit 103, the identification information assigning unit 103 assigns the identification information to the observation value of the second observing unit 102 at the time when the observation value of the first observing unit 101 is obtained. . Further, the identification information assigning unit 103 is configured to observe the first observation unit 101 based on the assignment of the identification information to the observation value of the second observation unit 102 at the time when the observation value of the first observation unit 101 is obtained. Among the possible combinations of assignment of identification information to the observation value of the second observation means 102 at a time when a value cannot be obtained, the amount of movement of the observation value assigned the identification information of the second observation means 102 for each observation time The combination of identification information is determined so that the sum of the two is minimized, and the identification information is also assigned to the observation value of the second observation means 102 at the time when the observation value of the first observation means 101 cannot be obtained. . Information obtained by the identification information assigning means 103 is output to the position estimating means 104.

  The position estimation unit 104 includes identification information and position information included in the observation value of the first observation unit 101, and identification information included in the observation value of the second observation unit 102 to which the identification information is allocated by the identification information allocation unit 103. The position of the moving object is captured by updating the estimated position of the observation object at each time when the observation value is obtained based on the position information.

  With this configuration, the identification information assigning unit 103 assigns identification information that is not included in the observation value of the second observation unit 102 based on the identification information of the observation value of the first observation unit 101. By configuring in this way, in the method of approximating identification information in the prior art, it is possible to avoid the problem that the identification information of a sensor that does not have identification information becomes ambiguous when the observed values are close to each other. . Therefore, the first embodiment has the problem of an increase in the position estimation error of the object due to the identification information becoming ambiguous, and further, an erroneous estimation of the movement path of the observation object due to the increase in the position estimation error. It can be avoided.

  Further, in the identification information assigning means 103, even when the observation frequency of the observation value of the first observation means 101 is lower than the observation frequency of the second observation means 102, the observation value of the second observation means 102 is changed to the observation value. Since the identification information can be assigned, it is possible to simultaneously avoid the problem that the information of the sensor with low observation frequency hardly works in the direction of correcting the position estimation error in the prior art.

  FIG. 2 is a block diagram showing a configuration of a human position tracking device 90 as a specific example of the moving object position estimating apparatus according to the first embodiment of the present invention. The human position tracking device 90 includes a wireless tag observation unit 201 as an example of a first observation unit, a human recognition camera 202 as an example of a second observation unit, and an ID allocation processing unit as an example of an identification information allocation unit 205 and a human position estimation unit 206 as an example of a position estimation unit. The human position tracking device 90 further includes an observation value acquisition unit 203, an observation value buffer 204, a human position estimated value recording unit 207, and a human position display unit 208.

  The wireless tag observation unit 201 receives radio waves transmitted from a wireless tag possessed by a person who is the object of observation with a plurality of antennas, and measures the position of the tag from the difference in arrival time for each antenna. At the same time, the wireless tag observation unit 201 combines the unique ID information assigned to each wireless tag with the information on the tag position, and outputs it to the observation value acquisition unit 203 as an observation value.

  The human recognition camera 202 discriminates between a person who is an observation target and a person other than the person, and outputs the position of the person existing in the imaging range to the observation value acquisition unit 203 as an observation value.

  FIG. 3 shows an example in which the wireless tag observation unit 201 and the human recognition camera 202 of the human position tracking device 90 in the first embodiment are arranged in the observation space 91. The person 89 to be observed possesses the wireless tags (wireless transmitters) 301, 302, and 303 so that the wireless tag observation unit 201 receives radio waves transmitted from the wireless tags 301, 302, and 303. The wireless tag observation unit 201 observes the positions and identification information of the wireless tags 301, 302, and 303. Further, the human recognition camera 202 is arranged by arranging the human recognition camera 202 from the ceiling toward the floor so that the imaging range of the human recognition camera 202 includes the entire floor surface 91f of the observation space 91 to be observed. Thus, the position of the person 89 moving on the floor surface 91f is observed.

  The observation value acquisition unit 203 has a timer 203 </ b> A that manages timing for acquiring observation values of the wireless tag observation unit 201 and the human recognition camera 202. The observation value acquisition unit 203 sets each observation value at the time interval set by the timer 203A as the observation cycle of the wireless tag observation unit 201 and at the time interval set by the timer 203A as the observation cycle of the human recognition camera 202. Acquired and additionally recorded in the observation value buffer 204.

  The observation value buffer 204 records the observation value acquired by the observation value acquisition unit 203 for each observation time. 4A and 4B show examples of observation values recorded in the observation value buffer 204. FIG.

  4A and 4B are examples in which there are three persons to be observed, and three wireless tag observation values and three human recognition camera observation values are recorded at each observation time. One observation value includes position information and ID information. The position information is coordinates where the person 89 exists in the observation space 91. In the first embodiment, an example in which the two-dimensional coordinates in the plane of the floor surface 91f of the space 91 where the person 89 exists is used as the position coordinates is shown. The ID information is information for distinguishing the observed persons 89 from each other. In the first embodiment, three people 89 are distinguished by ID information values of A, B, and C, respectively.

  The observation value recorded in the observation value buffer 204 is an observation value from the time when the RFID tag observation value is obtained to the time when the next RFID tag observation value is obtained. In the first embodiment, the observation period of the human recognition camera 202 is Δt, and the observation period of the wireless tag observation unit 201 is 4Δt. That is, the ratio of the observation frequencies of the wireless tag observation unit 201 and the human recognition camera 202 is assumed to be 1: 4. Accordingly, in FIG. 4A and FIG. 4B, observation values at five times from 0 to 4Δt are recorded.

  FIG. 4A shows a state immediately after the observation value acquisition unit 203 acquires observation values from the wireless tag observation unit 201 and the human recognition camera 202. The wireless tag observation value includes position information and ID information, but the human recognition camera observation value does not include identification information. For this reason, in FIG. 4A, the ID information of the observation value of the human recognition camera 202 is a value “UK” meaning “unknown”.

  FIG. 4B shows a state immediately after an ID assignment processing unit 205 (to be described later) assigns ID information to a human recognition camera observation value. In FIG. 4B, ID information assigned to the observation information ID information of the human recognition camera 202 is stored.

  The ID assignment processing unit 205 reads the content of the observation value recorded in the observation value buffer 204, the correspondence between the position information of the wireless tag observation value and the position information of the human recognition camera observation value, and the human recognition camera observation value Based on the evaluation of the movement amount of the observation value at each observation time, the assignment of the ID identification information to the human recognition camera observation value is determined, and the assigned ID identification information is used as the human recognition camera observation value in the observation value buffer 204. Write to ID information.

  The person position estimation unit 206 estimates the position of the person 89 at each observation time based on the observation value in the observation value buffer 204 in the state where the ID information is written. The person position estimation unit 206 holds an estimated position for each identified person. Since the correspondence between the observed value and the estimated position is immediately known from the identification information of the observed value in the observed value buffer 204, by updating the estimated position of the person 89 with the corresponding observed value at each observation time, the moving person 89 And the estimated value of the person position is recorded in the person position estimated value recording unit 207.

  In the human position estimated value recording unit 207, information on the estimated position of the person 89 estimated by the human position estimating unit 206 is recorded for the number of persons 89 at each observation time.

  FIG. 5 shows an example of the estimated position of the person 89 recorded in the person position estimated value recording unit 207. The estimated positions (two-dimensional coordinates) of the three persons A, B, and C who are distinguished by the ID at each observation time are recorded.

  The human position display unit 208 specifies a human position estimated value in an arbitrary time range of the estimated human position recorded in the human position estimated value recording unit 207, and further specifies a human position estimated value of an arbitrary person 89. Read and display the locus of each person's position. The human position display unit 208 is configured as a device including a display such as a CRT or a liquid crystal display.

  FIG. 6 shows a display example of the person position display unit 208. Here, in the first embodiment, three people (ID = A, B, C) 604, 605, and 606 are moving, and the arrow lines connecting between the circle marks indicated by 601 602, and 603 are shown. Represents the movement trajectory of each of the three persons 604, 605, 606. The estimated positions of the three persons 604, 605, and 606 from the latest observation value time (t = 4Δt) to the observation time seven times before are indicated by circles, and movement between each observation time is indicated by a circle. It is represented by a connecting arrow line.

  It should be noted that an image captured by the human recognition camera 202 in the direction from the ceiling to the floor is sent to the human position display unit 208, and the human position display unit 208 displays the video from the human recognition camera 202 on the movement locus 601 shown in FIG. , 602, 603 can be displayed in an overlapping manner.

  Reference numerals 604, 605, and 606 in FIG. 6 indicate people A, B, and C appearing in the video captured by the human recognition camera 202. Since the shooting is from the ceiling to the floor, the top and shoulders are mainly reflected.

  Next, based on the flowchart shown in FIG. 7, the flow of operation of the human position tracking device 90 in the first embodiment will be described.

  Steps S101 to S107 shown in FIG. 7 acquire observation values from the wireless tag observation unit 101 and the human recognition camera 102, respectively, and an ID assignment processing unit 103 assigns ID information to the human recognition camera observation values. This corresponds to an operation flow until the human position estimating unit 206 estimates the position of the person with respect to the observation value to which the ID information has been assigned.

  First, in step S101, the observation value acquisition unit 203 from the wireless tag observation unit 201 and the human recognition camera 202 at a time synchronized with the observation cycle set by the timer 203A in each of the wireless tag observation unit 201 and the human recognition camera 202. Get observations. Observation values are not acquired at times other than the time synchronized with the observation period. In the first embodiment, since the observation period of the wireless tag 201 is set to four times the observation period of the human recognition camera 202, when the wireless tag observation value and the human recognition camera observation value can be acquired simultaneously, Sometimes only human recognition camera observations can be acquired.

  Next, in step S <b> 102, the observation value acquisition unit 203 records the observation value acquired in step S <b> 101 in the observation value buffer 204.

  Next, in step S103, the ID assignment processing unit 205 determines whether or not the wireless tag observation value has been acquired in step S101. If the observation value acquisition unit 203 determines that the wireless tag observation value could not be acquired in step S101, the flow proceeds to step S104. In step S104, the process waits for Δt time, which is the observation cycle of the human recognition camera, and returns to step S101.

  If the observation value acquisition unit 203 determines in step S101 that the wireless tag observation value has been acquired, the process proceeds to step S105. In this case, similarly to the example shown in FIG. 4A, the observation value buffer 204 stores the RFID tag observation value and the human recognition camera observation value observed between the observation time at that time and the previous observation time. It is recorded.

  In step S <b> 105, the ID allocation processing unit 205 sets the human recognition camera observation value based on the position information and the ID information included in the wireless tag observation value and the human recognition camera observation value recorded in the observation value buffer 204. ID information assignment processing is performed. Details of the processing in step S105 will be described later.

  Next, in step S106, the human position estimation unit 206, the wireless tag observation value recorded in the observation value buffer 204, the position information of the human recognition camera observation value to which the ID information is assigned in step S105, and the ID Based on the information, the position of the person 89 is estimated. As a method for estimating the human position of the human position estimating unit 206, a previously proposed sequential estimation method can be used. For example, a Variant Data Association Filter (VADAF), which is a kind of online Bayes filter, can be used.

  Here, by performing position estimation processing based on the online Bayes filter, error characteristics of the position information included in the observation values of the RFID tag observation unit 201 and the human recognition camera 202, that is, variations in the observation values as the position sensors can be reduced. It is possible to perform position estimation in consideration, and it is possible to reduce an error from the position of the true person as compared with the position information of the observation value as it is as the position of the person. Even if the position information of the observation value is directly used as the position of the person, if the error does not become a problem, the position estimation process may not be performed, and the effect of the present invention is not impaired.

  Next, in step S <b> 107, the information on the human position estimated by the human position estimating unit 206 is written in the human position estimated value recording unit 207.

  The operation flow of the human position tracking device 90 has been described above.

  Next, details of the ID allocation processing by the ID allocation processing unit 205 in step S105 will be described.

  First, it will be described that the ID assignment process in step S105 can be formulated as a minimum cost route search problem which is a kind of combination search problem.

  After that, the A * algorithm can be applied as a procedure for solving the minimum cost route search problem, and the procedure for deriving the minimum cost route by the A * algorithm will be described.

  An explanation will be given of formulating the ID allocation process as a minimum cost route problem.

  FIG. 8 is a diagram illustrating examples of trajectories 801, 802, and 803 in which three persons 89 identified by the ID information of A, B, and C have actually moved in the following description of the formulation.

  FIG. 9 is a diagram illustrating an example of observation values obtained when three people move in the gray rectangular region 804 in FIG. 8. The observed values of the wireless tags of people A, B, and C are indicated by a black-filled square, a diagonally shaded square inside, and a square, respectively. The circles filled in with black are observation values of the human recognition camera 202.

In FIG. 9, the observation values included in the area 910 surrounded by the dotted rectangle are observation values obtained at the same observation time. Five dotted rectangular regions 910 are observation values obtained at the same time at each observation time from time t = 0 when the RFID tag observation value is obtained to time t = 4Δt when the next RFID tag observation value is obtained. . Here, assigning ID information to the human recognition camera observation values by the ID assignment processing unit 205 means that each of the observation values of the three human recognition cameras 202 at each observation time includes any ID information of A, B, and C. Is assigned by the ID assignment processing unit 205. In order to formulate this, the variable i is used as an observation value index to distinguish the three observation values at each observation time. i = 1 indicates the first observation value, i = 2 indicates the second observation value, and i = 3 indicates the third observation value. Next, the variable ID _i (t) is used to formulate the assignment of ID information to the i-th observation value. For example, ID _i (t) = A indicates that ID information A is assigned to the i-th observed value at time t.

Next, since the position information (coordinates) of the observation value is used for later formulation as a minimum cost path problem, x _i (t) is set as the coordinate of the i-th observation value at time t.

Regarding the formulation of the observed values and i, ID _i (t), x _i (t) described here, the correspondence relationship with the observed values is shown in FIG.

  FIG. 11 is a diagram illustrating a combination of ID information assignments to three human recognition camera observation values obtained when there are three persons. The combination of ID information of A, B, and C to the three observations is 3! It can be seen that there are six ways. For n people, n! Street.

  First, at the time t = 0 when the RFID tag observation value is obtained and the time t = 4Δt, it is possible to determine the optimum combination among the combinations of ID assignments to the observation values of these six types of human recognition cameras. it can. The sum of the distance between the position of the wireless tag observation value and the position of the observation value of the human recognition camera assigned the same ID information as the ID information of the wireless tag observation value (here, the sum of the three distances between the observation values) is the most. The ID allocation processing unit 205 selects a combination of ID allocations that is small. For example, at t = 0, the ID assignment processing unit assigns a combination of assigning ID information A to the observed value 901, assigning ID information B to the observed value 902, and assigning ID information C to the observed value 903 in FIG. Choose 205. At t = 4Δt, the ID assignment processing unit 205 may select a combination in which ID information B is assigned to the observed value 904, ID information C is assigned to the observed value 905, and ID information A is assigned to the observed value 906.

  Up to this point, it can be seen that one combination of ID assignments can be considered at time t = 0 and time t = 4Δt, and a combination of six ID assignments can be considered at other times.

  Here, the allocation state of ID information to the observation values of i = 1 to 3 is expressed in the order in which A, B, and C are arranged in three horizontally arranged rectangles.

  FIG. 12 is a diagram representing a search space of a combination of ID assignments to human recognition camera observation values from t = 0 to t = 4Δt by a search graph. The assignment state of the ID to the observation value of the human recognition camera at each observation time becomes a node of the search graph.

  Each time an observation time advances by one, an edge connecting nodes changes from an ID assignment state to an observation value represented by the node at which the edge begins to an ID assignment state to an observation value represented by the node at which the edge ends. Represents a transition. Each edge is assigned a cost value of Cost. The formula that defines Cost is listed below as (Formula 2).

(Equation 2)

  Cost (t, t + Δt) on the left side of (Equation 2) is assumed to have an ID assigned to each of n observation values at time t and n observation values at the next observation time t + Δt. In the ID allocation state, the distances between observation values having the same ID are totaled.

  In the search graph shown in FIG. 12, the path (minimum cost path) having the smallest cost among the paths traced from the node at t = 0 to the edge and the node at t = 4Δt is the ID allocation processing unit 205. Think in. The ID assignment to the observation value represented by the node existing on the route is the ID assignment with the smallest total movement distance of the observation values having the same ID between the observation times. That is, at time t = 0 and t = 4Δt, the ID is assigned to the optimum human recognition camera based on the positional relationship with the ID information of the RFID tag observation value, and the time t = 0 while satisfying the restriction. The ID assignment processing unit 205 obtains the optimum ID assignment from the viewpoint of the moving distance to the observation value of the human recognition camera other than t = 4Δt. Therefore, the ID assignment problem to the human recognition camera observation value can be formulated as a minimum cost route search problem in the ID assignment processing unit 205.

  Next, it will be described that the ID assignment processing unit 205 can apply the A * algorithm as a procedure for solving the minimum cost route search problem formulated so far. The A * algorithm is known as a kind of algorithm that solves the search problem based on the minimum cost. By introducing the prediction cost into the search procedure, it is more efficient than the method that does not depend on the prediction cost such as Dijkstra method. In particular, the ID allocation processing unit 205 can derive the minimum cost route.

Hereinafter, the A * algorithm will be described. As shown in FIG. 12, assuming that a cost is set for each edge of the search graph, and a cost of a route that passes through an arbitrary node n in the search graph and is the minimum cost is f (n), f (n) is
f (n) = g (n) + h (n)
It can be expressed as. g (n) is the minimum cost from the start node to n, and h (n) is the minimum cost from n to the target node. The route that passes through the node n and has the lowest cost is unknown, and g (n), h (n), and f (n) are also unknown. Therefore, an estimated value f * (n) of f (n) is considered.

f * (n) = g * (n) + h * (n)
Here, g * (n) is an estimated value of the minimum cost from the start node to n, and h * (n) is an estimated value of the minimum cost from n to the target node. g * (n) can be estimated from the integration of the cost assigned to the edge by sequentially proceeding with the search from the start node, but h * (n) cannot be obtained during the search. Therefore, by giving an appropriate estimate for h * (n), evaluating the estimated cost f * (n) of the route through node n, and following the route that gives a smaller f * (n), The A * algorithm searches for the least cost path.

h * (n) is called the heuristic function of the A * algorithm, and for all n, h * (n) is
h * (n) ≦ h (n)
When satisfying, the path from the start node to the target node determined by the A * algorithm is guaranteed to be the minimum cost path. Conversely, if h * (n) satisfying this equation can be set, the problem can be solved by the A * algorithm.

In the first embodiment, a method for setting h * (n) will be described. Let n be one of the nodes in the ID assignment state at time Δt. As h * (n), as indicated by a broken line in FIG. 13, the total distance between observation values having the same ID from the observation value at time Δt to the observation value at time 4Δt is considered as h * (n). The ID allocation processing unit 205 immediately knows that the distance of the broken line is the shortest among the distances from the observed value at time t to the observed value at time 4Δt for the observed value of a certain ID. , For true minimum cost h (n) from n to the target node at time 4Δt,
h * (n) ≦ h (n)
Holds. Therefore, the ID assignment processing unit 205 can solve the ID assignment problem of the first embodiment by the A * algorithm.

  Next, a procedure for solving the ID assignment problem by the A * algorithm in the ID assignment processing unit 205 will be described. FIG. 14 is a diagram illustrating a node expression format when the ID assignment processing unit 205 solves with the A * algorithm. One node consists of a time t, an ID assignment state to each observation value, an estimated cost f * value of the node, and a pointer to the previous node. By following the pointer to the node, it is possible to know the node that has passed from the start node. FIG. 14 shows an example in which the pointer of the node at time Δt has a pointer to the node at time 0 (start node).

  FIG. 15A and FIG. 15B are flowcharts of procedures when the ID assignment processing unit 205 solves with the A * algorithm. 15A and 15B is executed by the ID assignment processing unit 205.

  Ts and te in the flow of FIG. 15A are times when the RFID tag observation values are obtained, respectively, and ts <te.

  Steps S <b> 201 to S <b> 203 are processes in which the ID assignment processing unit 205 sets a start node for an ID assignment search problem.

  First, in step S201, the ID assignment processing unit 205 obtains a correspondence that minimizes the distance between the wireless tag observation value and the human recognition camera observation value at time ts when the wireless tag observation value is obtained.

  Next, in step S202, based on the correspondence, ID information of the wireless tag observation value corresponding to the human recognition camera observation value is assigned by the ID assignment processing unit 205.

  In step S203, the ID assignment processing unit 205 creates a start node from the assignment state.

  Steps S <b> 204 to S <b> 206 are processing for setting a target node for an ID assignment search problem in the ID assignment processing unit 205.

  In step S204, the ID assignment processing unit 205 obtains a correspondence relationship that minimizes the distance between the wireless tag observation value and the human recognition camera observation value at time te when the wireless tag observation value is obtained.

  Next, in step S205, based on the correspondence, ID information of the RFID tag observation value corresponding to the human recognition camera observation value is assigned by the ID assignment processing unit 205.

  In step S206, the ID assignment processing unit 205 creates a target node from the assignment state.

  Next, the procedure following step S206 will be described with reference to the flowchart of FIG. 15B.

  In FIG. 15B, s is a start node, and the OPEN list and the CLOSE list are lists for storing nodes, respectively. The CLOSE list is a list that stores nodes that have been searched for the node ahead of the own node, and the OPEN list is a list that stores nodes that have not been searched for from the own node to the next node.

  Step S207 is the start of the A * algorithm, and the ID assignment processing unit 205 puts the start node in the OPEN list. The ID allocation processing unit 205 sets the estimated cost f * (s) of the start node at this time to f * (s) = h * (s). In the first embodiment, it is the sum of the distances between the observed values of the same IDs of the observed value of the start node and the observed value of the target node.

  In step S208, the ID assignment processing unit 205 checks whether the OPEN list is empty. If the ID assignment processing unit 205 determines that the OPEN list is empty, the process ends. When the OPEN list is empty, this corresponds to a case where a route reaching the target node has not been found. However, in the search problem of the first embodiment, the route from the start node to the target node is always the ID allocation processing unit 205. Can be found at When the ID assignment processing unit 205 determines that the OPEN list is not empty, the process proceeds to step S209.

  Next, in step S209, the ID assignment processing unit 205 extracts the node n having the smallest value of f * from the OPEN list and moves it to the CLOSE list.

  Next, in step S210, the ID allocation processing unit 205 determines whether n is a target node, and if the ID allocation processing unit 205 determines that n is a target node, the search ends. At this time, the path of the node obtained by tracing the pointer of n until it reaches the start node is the minimum cost path. When the ID assignment processing unit 205 determines that n is not the target node, the process proceeds to step S211.

  Next, in step S211, the ID assignment processing unit 205 generates a hypothesis of the ID assignment state at the observation value time next to the observation time of the n node, and sets these as a node M as a continuation node of the node n. For each node m of M, the ID assignment processing unit 205 calculates the value of f ′ (m) = g * (n) + h * (m) + Cost (n, m). Cost (n, m) is a cost assigned to the edge from the node n to the node m, and is calculated by the ID assignment processing unit 205 based on (Equation 2).

  In step S212, the ID assignment processing unit 205 extracts one node from M and sets it as m.

  Next, in step S213, in the ID assignment processing unit 205, the node m matches the node included in the OPEN list, or the node m matches the node included in the CLOSE list, or the node in either of the lists. Determine if they do not match.

  When the ID assignment processing unit 205 determines that the node m is not included in either list, the process proceeds to step S214. In step S214, the ID allocation processing unit 205 sets f ′ (m) calculated in step S211 as the predicted cost f * of node m, sets the pointer of node m to node n, and adds it to the OPEN list.

  When the ID assignment processing unit 205 determines that the node m is included in the OPEN list, the process proceeds to step S215. In step S215, the ID allocation processing unit 205 compares the predicted cost f * (m) of the included node with f ′ (m) calculated in step S211, and f ′ (m) <f * (m). Then, the value of f * is replaced with f * (m) = f ′ (m), and the pointer of node m is set to node n.

  When the ID assignment processing unit 205 determines that the node m is included in the CLOSE list, the process proceeds to step S216. In step S216, the ID allocation processing unit 205 compares the predicted cost f * (m) of the included node with f ′ (m) calculated in step S211, and f ′ (m) <f * (m). Then, the value of f * is replaced by f * (m) = f ′ (m), the pointer of node m is set to node n, and m is moved from the CLOSE list to the OPEN list.

  Following step S214, step S215, and step S216, in step S217, the ID assignment processing unit 205 determines whether all m have been extracted from the node set M. If the ID assignment processing unit 205 determines that all m has not been extracted, the process returns to step S212. If the ID assignment processing unit 205 determines that all m have been extracted, the process returns to step S208.

  In step S210, the ID allocation processing unit 205 displays the ID allocation state expressed by the node between the target node and the start node obtained by following the pointer of the target node obtained when n reaches the target node and ends. It is an ID assignment state to the human recognition camera observation value in each observation value to be obtained.

  Now, in step S211 of the solution procedure in the A * algorithm, when the ID allocation state hypothesis at the next observation time of the node n is generated by the ID allocation processing unit 205, all the possible ID allocations have been described so far. The state is considered by the ID allocation processing unit 205. For example, when there are three observation values, the ID assignment processing unit 205 generates all six assignment states as hypotheses.

  However, if the observation target of the first embodiment is a person, there is a limit to the distance that the person can actually move during the observation period Δt. Therefore, when the ID assignment processing unit 205 generates the ID assignment state hypothesis in step S211, the hypothesis considered as the ID assignment state from the ID assignment state of the node n to the observation value at the next observation time is included. May also include hypotheses that are not feasible due to human movement. Since such a hypothesis cannot be realized by human movement, it is useless to generate and search. In FIG. 16A, the distance that the observation value of the same ID moves from the ID assignment of the observation value at time 0 to the ID assignment at time Δt is all within the movable range (the range of the dotted circle in the figure). An example is shown.

  On the other hand, FIG. 16B shows the range in which the observation value of the same ID moves from the ID assignment of the observation value at time 0 to the ID assignment at time Δt, and B is a human movable range (the dotted line in the figure). In this example, A and C exceed the human movable range (the dotted circle in the figure).

  In the first embodiment, the setting of the range in which a person can move from a certain position (the range of the dotted circle in the figure) is the maximum measured in advance in the observation period Δt and the observation target space with the position as the center. It is assumed that the distance calculated by the product of the movement speeds of the person is the inside of a circle having a radius.

  Of course, it is possible to set the movable range of the person by other methods. For example, the predicted position of the observed value at the observation time can be set based on the moving speed and moving direction expected from the observed movement history.

  Therefore, the hypothesis of ID assignment at time Δt in FIG. 16A can be generated, but the hypothesis of ID assignment at time Δt in FIG. 16B cannot be generated.

  Accordingly, in step S211, the search space is reduced by discarding the hypothesis that the movement distance of the observed value of the same ID exceeding the limit of the human movable distance occurs as shown in the example of FIG. The search can be performed efficiently.

  With this configuration, the ID assignment processing unit 203 assigns ID information to the observation value of the human recognition camera 202 that does not have ID information based on the ID information of the wireless tag observation unit 201, so that the observation objects approach each other. It is possible to determine the ID information of the observation value in any situation including the situation where the estimation error occurs in the situation where the error of the human position estimation including the passing of people is likely to occur as a result. The problem can be solved. As an example, FIG. 28 shows an example of a person position movement path estimation result using the first embodiment when a person moves along the movement paths 1702 and 1703 as in FIG. From FIG. 28, it can be seen that the human position movement path can be estimated in substantially the same manner as the human movement paths 1702 and 1703 in FIG.

  In the first embodiment, the estimated position of the person position is displayed on the person position display unit 208 such as a display. However, the estimated form of the person position can be used as an input to another apparatus that requests the position of the person, and does not limit the scope of the present invention.

(Second Embodiment)
FIG. 20 is a block diagram showing a configuration of a human position tracking device 90A in the second embodiment of the present invention. The second embodiment of the present invention is configured to perform ID assignment processing for human recognition camera observation values even when the number of human recognition camera observation values does not match the number of RFID tag observation values due to concealment or misrecognition. It is a thing. In the second embodiment of the present invention, the ID allocation processing unit 205 of the first embodiment of the present invention is replaced with an ID allocation processing unit 210 having a different internal configuration. The same as in the first embodiment. The description of the components already described in the description of the first embodiment is omitted.

  The ID allocation processing unit 210 is configured to include an ID allocation control unit 210A, an observation value movement possibility determination unit 210B, and a virtual observation value generation unit 210C as internal configurations.

  The ID allocation control unit 210A controls the entire processing of the ID allocation processing unit 210. The ID allocation control unit 210A generates a search graph of hypotheses for ID allocation for a plurality of observation values of the human recognition camera 202 at each observation time, as with the ID allocation processing unit 205 of the first embodiment of the present invention. After that, by deriving the route with the lowest cost on the search graph, the optimum assignment of ID information to the human recognition camera observation value at each observation time is determined.

  However, if the number of human recognition camera observations does not match the number of RFID tag observations at each observation time, or observation is made at a position expected from the movement capability or movement history of the moving object (eg, person) to be observed Even if there is no value or there is an observation value at a position deviating from the expected position, in order to generate a hypothesis node for ID assignment, the observation value movement possibility determination unit 210B has one unit. After determining the combination of observation values that can be moved between the observation times, if there is no observation value to which an ID should be assigned, the virtual observation value generation unit 210C has a virtual position at the position expected from the observation history of the observation value. Generate observations and allow all IDs to be assigned.

  Next, an operation flow of the human position tracking device 90A in the second embodiment will be described. FIG. 21 is a flowchart showing an operation flow of the human position tracking device 90A of the second embodiment.

  The operation of the second embodiment is basically the same as the operation flow of the human position tracking device 90 in the first embodiment, except for the processing content in step S105A, The configuration of the operation flow is the same as the operation flow shown in the flowchart of FIG.

  As described above, step S105A differs from step S105 in that the processing content includes a response to a case where the number of RFID tag observation values does not match the number of human recognition camera observation values. Specifically, in the process of assigning IDs to human recognition camera observation values, the process of generating a hypothesis node for ID assignment to human recognition camera observation values is different, so the operation flow will be described focusing on this part. To do.

In the observation value buffer 204, as shown in FIG. 4, human recognition camera observation values are obtained at an observation interval Δt. Generating Here, the ID assignment state to human recognition camera observations in observation time t = t _0, a hypothesis ID assignment state to human recognition camera observations in observation time t = t 0 ₊ Δt with ID allocation processing unit 210 Think about the situation.

  FIG. 22 is a flowchart of an operation of generating an ID assignment hypothesis node for the human recognition camera observation value in this situation by the ID assignment processing unit 210.

First, in step S2201, the observed value movability determination unit 210B is movable to each observation of the human recognition camera 202 at time _t = t ₀ + Δt from each observation of the human recognition camera 202 at time t = _{t 0} Determine sex. FIG. 23 is an example showing movement between observation values from observation time t = t ₀ to observation time t = t ₀ + Δt. The numbers on the observation value side are numbers for distinguishing observation values at each observation time.

A circle 2301 indicated by a dotted line centered on the observation values 1 to 5 at the observation time t = t ₀ in FIG. 23 allows a person (moving body) to move during the observation time interval Δt. Represents the boundaries of various ranges. In the second embodiment, the designation of the range in which the observation target person 89 can move is determined from the observation time interval Δt and the limit of the moving speed of the observation target person 89, as in the first embodiment of the present invention. It is within a circle 2301 having a radius.

The criterion of the movement possibilities, the observation value movability determination unit 210B, each observed value definitive measurement time t = t ₀ is determined whether movable to a position of the observation time t = t 0 ₊ Δt observations of the throat To do.

FIG. 24 is a table showing the movement possibility determination result of the observed value movement possibility determination unit 210B. The table 2401 on the left side of FIG. 24 shows the result of the movement possibility determination in step S2201 in the form of a table. In Table 2401, and the vertical axis the identification number of observations at the observation time t = t _0, the horizontal axis is the identification number of observations at the observation time t = t 0 ₊ Δt. From the position of the observed value in the observation time t = t _0, if it is movable to a position of the observation value at the observation time t = t 0 ₊ Δt are indicated by "○", if it is not movable are indicated by "×" .

Next, in step S2202, the ID assignment control unit 210A performs observation that cannot be moved from any observation value at observation time t = t ₀ based on the observation value movement possibility determination result obtained in step S2201. The observed value at time t = t ₀ + Δt is excluded from ID assignment targets.

In the determination result of the observation value movement possibility shown in FIG. 24, the observation value with the identification number 3 at the observation time t = t ₀ + Δt cannot be moved from any observation value at the observation time t = t ₀ . The observation value is excluded from the ID allocation target by the ID allocation control unit 210A.

Next, in step S2203, the virtual observation value generation unit 210C performs the observation for the observation value at observation time t = t ₀ where none of the observation values at observation time t = t ₀ + Δt is movable. A virtual observation value is generated at a movement position expected from the movement history of the value.

In movability determination result of the observations shown in Figure 24, the observed value of the identification number 5 in the observation time t = t _0, the observation time t = t 0 ₊ Δt not be movable in the observations of the throat. Therefore, in the virtual observation value generation unit 210C, the movement predicted position of the observation value at the observation time t = t ₀ + Δt is obtained from the movement history of the observation value, a virtual observation value is generated at the position, and the observation value buffer 204 is generated. Add the information of the generated virtual observation value.

  FIG. 25 is a diagram showing the observed value at the observation time and the generated virtual observed value (O assigned the identification number 5 * in FIG. 25) in addition to the observed value at the observation time shown in FIG. Since the observation value with the observation time identification number 5 is generated by the virtual observation value generation unit 210C, the hypothesis that the observation value with the observation time identification number 5 has moved (that is, the hypothesis that the same ID is assigned to both) is generated. In the virtual observation value generation unit 210C, based on the hypothesis, it is assumed that the movement speed and the movement direction from the observation time to the observation time are maintained, and the movement predicted position at the observation time is represented by “◯” in FIG. And a virtual observation value (indicated by “◯” in the figure) is generated there.

  Next, in step S2204, the observation value movement possibility determination unit 210B re-evaluates the movement possibility with respect to the virtual observation value at the observation time generated by the ID assignment virtual observation value generation unit 210C. The result of re-evaluation by the observation value movement possibility determination unit 210B is shown in a table 2402 on the right side of FIG. The virtually generated observation value is distinguished by an identification number in the form of * added to the next number of the number of observation values that are truly obtained (if there are multiple virtual observation values, the number is added sequentially).

  Next, in step S2205, the ID assignment control unit 210A responds to the movement of the observation value at the observation time and the observation value at the observation time obtained as a result of determining the movement possibility of the observed observation value by the processing up to step S2204. Based on the relationship, a plurality of hypotheses are generated by assigning the ID assigned to the observation value at the observation time to the observation value at the observation time in the correspondence relationship.

  FIG. 26 is a table showing combinations of hypotheses for assigning IDs to observed values obtained from the result of determining the possibility of movement of observed values in FIG. In the ID allocation control unit 210A, ID allocation that allows multiple allocation of IDs is performed on the observation values of the identification numbers 1 and 2 among the observation values at the observation time. ID = D and ID = E are assigned to the identification number 4 and the identification number 5 * of the observation time by the ID assignment control unit 210A, respectively.

  FIG. 27 is a diagram showing the representation of hypothesis nodes corresponding to the three ID assignment hypotheses shown in FIG. The internal structure of the hypothesis node is the same as that of the hypothesis node described in FIG. 14 of the first embodiment. When two or more IDs are assigned to one observation value, “′” (dash) is added to the identification number of the observation value to indicate that a plurality of IDs are assigned to the same identification number. In the case of ID assignment hypothesis 1, since two IDs of ID = B and ID = C are assigned to the observed value of identification number 2, multiple IDs are assigned to ID = C and recorded in the node as 2 ′. To do. When two or more IDs are assigned, “′” is sequentially increased to distinguish them.

  With this configuration, within the ID assignment processing unit 210, the observation value movement possibility determination unit 210B is configured to detect an error in the observation value due to an observation error and an error based on the person's movement possibility in one unit of observation interval. While identifying the state of the erroneous observation value by detection, the correspondence between the movable observation values is determined, and the virtual observation value generation unit 210C is also predicted based on the observation value movement history regarding the disappearance of the observation values. By virtually generating an observation value at the movement destination position, it is possible to realize a human position tracking device 90A that can cope with an observation error in the human recognition camera 202.

  In the second embodiment, the estimated position of the person position is displayed on the person position display unit 208 such as a display. However, it is of course possible to use the human position estimation utilization form as an input to another apparatus that requests the human position, and does not limit the scope of the present invention.

  In the first embodiment and the second embodiment, the first observation unit 101 receives radio waves transmitted from the wireless tags 301 to 303 including the identification information of the moving body (for example, a person) and held by the moving body. The wireless tag observation unit 201 outputs body identification information and position information as observation values. However, the present invention is not limited to this. The ultrasonic wave is transmitted to the moving body, and the reflected ultrasonic wave is received and moved. It can also be constituted by an ultrasonic sensor that outputs body identification information and position information as observation values.

  In the first embodiment and the second embodiment, the second observation unit 102 is configured by a camera 202 that captures an image and outputs position information of a moving body (for example, a person) as an observation value. Not limited, Millimeter wave transmitter / receiver that transmits millimeter wave to mobile body and receives reflected millimeter wave and outputs the position information of mobile body as observation value, or multiple on the floor of observation space It can also be configured by a pressure sensor that outputs position information of the moving object as an observation value.

  In addition, it can be made to show the effect which each has by combining arbitrary embodiment or modification of the said various embodiment or modifications suitably.

  A mobile object position estimation device, a mobile object position estimation method, and a mobile object position estimation program according to the present invention are a system for recording and monitoring the behavior of a person to be observed or an analysis of a worker's movement history and The system can be applied to a system for evaluating efficiency and a system for tracing the position of a package at a distribution site.

89 people 90, 90A human position tracking device 91 observation space 91f floor 101 first observation means 102 second observation means 103 identification information assignment means 104 position estimation means 201 wireless tag observation part 202 human recognition camera 203 observation value acquisition part 204 observation Value buffer 205 ID allocation processing unit 206 Human position estimation unit 207 Human position estimation unit Recording unit 208 Human position display unit 210 ID allocation processing unit 210A ID allocation control unit 210B Observation value mobility determination unit 210C Virtual observation value generation unit 300 Room 301 Wireless Tag 302 Wireless Tag 303 Wireless Tag 401 Example of Observation Value Recorded in Observation Value Buffer 402 Example of ID Assignment Process Performed on Observation Value Recorded in Observation Value Buffer 601 Human Trajectory of Person Position Display Unit Display example 602 Display of the movement locus of the person in the person position display section Example 603 Display example of movement trajectory of person on person position display section 604 Image captured from person's head direction displayed on person position display section 605 Image captured from person's head direction displayed on person position display section 606 Image of the person from the top of the head displayed on the human position display unit 901 Example of human recognition camera observation value 902 Example of human recognition camera observation value 903 Example of human recognition camera observation value 904 Example of human recognition camera observation value 905 Example of human recognition camera observation value 906 Example of human recognition camera observation value 1701 Example of movement path of an object for explaining a problem in the prior art 1701A An object for explaining a problem in the prior art approaches Regions 1702 and 1703 Movement path 2301 Boundary 2401 of the range in which a person (moving body) can move during each observation time interval Δt. Observation result determination result 2402 Virtual observation value The determination result of the movement possibilities against

Claims (12)

Observation of a plurality of moving objects existing in the observation space, and input the observation value including the identification information of the moving object and the position information of the moving object obtained sequentially for each moving object, the plurality of movements to be observed A body position estimation device for estimating body identification information and position information,
First observation means for outputting identification information and position information of the mobile body as observation values at predetermined time intervals;
Second observation means for outputting position information of the moving body as an observation value at a time interval different from the interval of the first observation means;
Position information of the observed value of the first observing means and the observation of the second observing means in an observation section of the time interval of each of the first observing means and the second observing means. The total of the relative distance of the position information of the value and the total amount of movement of the observation value of the second observation unit in the observation section are minimized with respect to the observation value of the second observation unit. Identification information assignment means for assigning identification information obtained from the first observation means;
Position estimation means for estimating the position of the moving object based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means The
A moving body position estimation apparatus having the same. The first observation means receives the radio wave that includes the identification information of the mobile body and is transmitted from the radio tag of the mobile body, and outputs the identification information and position information of the mobile body as observation values Or an ultrasonic sensor that receives ultrasonic waves reflected by transmitting ultrasonic waves to the moving body and outputs identification information and position information of the moving body as observation values,
The second observing means captures an image and outputs position information of the moving body as an observation value, or receives a millimeter wave transmitted from the moving body and reflected from the moving body, 2. The movement according to claim 1, wherein the movement is a millimeter wave transmitter / receiver that outputs position information as an observation value, or a pressure sensor that is arranged on a floor surface of the observation space and outputs position information of the moving body as an observation value. Body position estimation device.   In the identification information allocating means, the observation value of the second observing means outside the movable range in which the moving body can move within the time of the observation interval is deleted, and the moving body within the time of the observation interval. Based on the observed value of the second observing means within the movable range, the identification that is not movable from the position of the previous observed value of the second observing means when performing the identification information assignment process Specifying an observation value when performing information allocation, or when there is no one that can be moved to the position of the observation value of the second observation means when performing identification information allocation, when performing identification information allocation The mobile position estimation according to claim 1 or 2, wherein a previous observation value is specified, and identification information obtained from the first observation means is assigned to the specified observation value of the second observation means. apparatus.   The identification information allocating means is a previous observation when the identification information allocation is performed, which is specified that there is no object that can be moved to the position of the observation value of the second observation means when the identification information allocation is performed. A virtual observation value generation unit that predicts a movement destination position when assigning the identification information to the value based on the movement history of the observation value and generates a virtual observation value at the prediction position; The moving body position estimation apparatus according to claim 3.   In the identification information allocating means, the movable range of the mobile object to be observed is a circle range having a radius that is a product of the maximum moving speed based on the measurement result of the moving ability of the mobile object and the observation interval. The mobile body position estimation apparatus according to claim 3 or 4, which is set as follows.   In the identification information allocating means, the movable range of the observation target moving body is defined as the position of the previous observed value with respect to the time when the identification information is allocated in the movement history of the observed value of the second observing means. And a circle having a predetermined radius centered on the movement destination position obtained on the assumption that the movement speed and movement direction calculated from the position of the observation value two times before the time when the identification information is assigned are maintained. The moving body position estimation apparatus according to claim 3 or 4. Observation of a plurality of moving objects existing in the observation space, and input the observation value including the identification information of the moving object and the position information of the moving object obtained sequentially for each moving object, the plurality of movements to be observed A body position estimation method for estimating body identification information and position information,
The identification information and position information of the mobile object are output as observation values by the first observation means at predetermined time intervals,
The position information of the moving body is output as an observation value by the second observation unit at a time interval different from the interval of the first observation unit,
Position information of the observed value of the first observing means and the observation of the second observing means in an observation section of the time interval of each of the first observing means and the second observing means. The total of the relative distance of the position information of the value and the total amount of movement of the observation value of the second observation unit in the observation section are minimized with respect to the observation value of the second observation unit. The identification information obtained from the first observation means is assigned by the identification information assignment means,
Based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means, the position of the moving body is estimated by the position estimation means To
Mobile object position estimation method. A first observation value including identification information and position information of the moving object and output from the first observation means; and a second observation value including position information of the moving object and output from the second observation means. From the first time when the first observation value was obtained at the same time, the second observation obtained between the second time when the first observation value and the second observation value were obtained simultaneously. For value
At the first time, the identification information assigning means obtains a combination having the shortest distance between the position of the first observation value and the position of the second observation value,
Based on the combination at the first time, the identification information of the first observation value is allocated by the identification information allocation means as the identification information of the second observation value,
Further, at the second time, the identification information assigning means obtains a combination having the shortest total distance between the position of the first observation value and the position of the second observation value,
In the combination at the second time, the identification information of the first observation value is assigned by the identification information assigning means as the identification information of the second observation value,
In the identification information allocating means, the allocation state of the identification information to the second observation value at the first time is a start node, and the identification information to the second observation value at the second time is A node that is located between the start node and the target node, with a combination of identification information assigned to the second observation value between the first time and the second time, with the assignment state as the target node In the search graph obtained by setting the transition cost between the nodes as the total movement amount of the second observation value, the minimum cost path from the start node to the target node of the search graph is determined. By obtaining, the identification information allocating means assigns the identification information of the first observation value to the second observation value from the first time to the second time.
The moving body position estimation method according to claim 7.   In the search process on the search graph, when the identification information assigning unit generates a node corresponding to a hypothesis of an assignment state at the next observation time from a node corresponding to the assignment state of the identification information at a certain observation time. , Taking into account the movable range of the mobile object to be observed, determining the correspondence of the observation values of the second observation means movable within the time of the observation interval, An observation value that is not movable from the position of the observation value of the second observation means and that is not movable from the position of the previous observation value when performing the search process by specifying the observation value when the search process is performed The mobile object position estimation method according to claim 8, wherein an assignment hypothesis node including a node is not generated.   In the search process on the search graph, when the identification information assigning unit generates a node corresponding to a hypothesis of an assignment state at the next observation time from a node corresponding to the assignment state of the identification information at a certain observation time. In consideration of the movable range of the mobile object to be observed, the previous observation for the time when the search process was performed, when it was specified that there was no object that could be moved to the position of the observation value when the search process was performed Predicting a destination position when performing the search processing based on a movement history of the observed value, generating a virtual observed value at the predicted position, and adding the virtual observed value to the virtual observed value. The moving body position estimation method according to claim 8 or 9, wherein a hypothesis node that the observed value has moved is generated. Observation of a plurality of moving objects existing in the observation space, and input the observation value including the identification information of the moving object and the position information of the moving object obtained sequentially for each moving object, the plurality of movements to be observed A body position estimation program for estimating body identification information and position information,
First observation means for outputting the identification information and position information of the mobile object as observation values at predetermined time intervals, and a time different from the interval of the first observation means using the position information of the mobile object as observation values Position information of the observed value of the first observing means and position of the observed value of the second observing means in a certain observation section of each time interval with the second observing means outputting at intervals The first relative to the observation value of the second observation means is such that the total of the relative distance of information and the total amount of movement of the observation value of the second observation means in the observation section are minimized. Identification information assigning means for assigning identification information obtained from the observation means;
Position estimation means for estimating the position of the moving object based on the observation value of the first observation means and the observation value of the second observation means including the identification information assigned by the identification information assignment means A moving body position estimation program for causing a computer to realize the function. A first observation value including identification information and position information of the moving object and output from the first observation means; and a second observation value including position information of the moving object and output from the second observation means. From the first time when the first observation value was obtained at the same time, and then the second time obtained during the second time when the first observation value and the second observation value were obtained simultaneously. For observations,
A function for obtaining a combination of the shortest distance between the position of the first observation value and the position of the second observation value by the identification information assigning means at the first time;
A function of assigning identification information of the first observation value as identification information of the second observation value by the identification information assigning means based on the combination at the first time;
A function for obtaining a combination of the shortest distance between the position of the first observation value and the position of the second observation value by the identification information assigning unit at the second time;
A function of assigning the identification information of the first observation value by the identification information assigning means as the identification information of the second observation value in the combination at the second time;
The assignment state of identification information to the second observation value at the first time is a start node, and the assignment state of identification information to the second observation value at the second time is a target node, Search obtained by setting a combination of assignment of identification information to the second observation value between the first time and the second time as a node located between the start node and the target node In the graph, by setting the transition cost between the nodes as the total amount of movement of the second observation value, the minimum cost path from the start node to the target node of the search graph is obtained, thereby obtaining the first cost. A function of assigning identification information of the first observation value to the second observation value from the time of the second time to the second time by the identification information assigning means;
The moving body position estimation program according to claim 11, for causing a computer to realize the above.

Images