Tracking method, tracking device, and tracking program

Abstract

This tracking device executes processing of: acquiring a tracking result that is a result of detection and tracking of a subject by each of cameras included in a set having at least some of a plurality of cameras as components, the components of the set being changed by movement of each of the plurality of cameras; and selecting, as a representative tracking result, one tracking result satisfying a criterion from among the tracking results.

Description

Tracking Method, Tracking Device, and Tracking Program

[0001] The present invention relates to a tracking method, a tracking device, and a tracking program.

[0002] In recent years, connected cars equipped with cameras and sensors have become widespread. Also, as a technology related to digital twins, a technology that maps the positions of vehicles, people, accidents, etc. to world coordinates in real time based on information obtained from connected cars is expected.

[0003] In order to realize such a technology, a tracking technology for a subject using a plurality of cameras is a prerequisite. Conventionally, a technology for tracking a subject using a plurality of cameras is known (see, for example, Non-Patent Document 1).

[0004] Wenqian Liu, Octavia Camps, Mario Sznaier, "Multi-camera Multi-object Tracking", [online], [searched on September 9, 2024], Internet (https: / / arxiv.org / pdf / 1709.07065)

[0005] However, in the conventional technology, there may be a large deviation between the tracking result and the fact occurring in the real space.

[0006] For example, when tracking the same subject with a plurality of cameras, the identification of the target may be incorrect. For example, when observing one person with three cameras, if each camera identifies the person being tracked as a different person, while the actions occurring in the real space are continuous actions performed by the same person, the output of the tracking result will be non - continuous actions performed by three people.

[0007] Furthermore, when tracking the same subject with multiple cameras, it is possible that the subject may be missed or / or replaced. For example, when a fixed camera detects a dangerous vehicle (e.g., a blacklisted vehicle, a speeding vehicle, etc.) and requests video from a nearby connected car to continue tracking even outside the field of view, it is possible that a vehicle of the same make or color is traveling at the same speed behind the dangerous vehicle. In such cases, the system may end up tracking the other vehicle instead of the dangerous vehicle that was the target of the tracking.

[0008] Conventional technologies have not solved the problems that arise when tracking the same subject with multiple cameras, as described here.

[0009] The present invention has been made in view of the above, and aims to reduce the discrepancy between the tracking results and the facts occurring in real space.

[0010] To solve the above-mentioned problems and achieve the objective, the tracking method according to the present invention is characterized in that a computer performs a process of acquiring tracking results, which are the results of each camera in a set comprising at least some of a plurality of cameras as constituent elements, wherein the constituent elements change as each of the plurality of cameras moves, and selecting one tracking result that satisfies the criteria from among the tracking results as a representative tracking result.

[0011] According to the present invention, it is possible to reduce the discrepancy between the tracking results and the facts occurring in the real world.

[0012] Figure 1 is a diagram showing an example of the server configuration according to the first embodiment. Figure 2 is a flowchart showing the flow of pipeline processing and integration processing according to the first embodiment. Figure 3 is a diagram illustrating the data flow related to integration processing. Figure 4 is a diagram illustrating an overview of integration processing. Figure 5 is a diagram illustrating an example of integration processing. Figure 6 is a diagram showing an example of a computer that executes a tracking program.

[0013] Hereinafter, one embodiment of the present invention will be described in detail with reference to the drawings. However, the present invention is not limited by this embodiment. Furthermore, in the drawings, the same parts are denoted by the same reference numerals.

[0014] [First Embodiment] The server configuration will be explained using Figure 1. Figure 1 is a diagram showing an example of the server configuration according to the first embodiment. Here, the server performs tracking-related processing based on information acquired from multiple cameras. The server is an example of a tracking device. For example, tracking is the process of identifying the time-series positional information of a subject captured by a camera.

[0015] Furthermore, in this embodiment, the multiple cameras include surveillance cameras installed on the road and cameras mounted on the connected car. The subjects of tracking are people and vehicles on and around the road. However, the camera configuration and subjects are not limited to those described herein.

[0016] As shown in Figure 1, the server 10 has an integration unit 20 and an inference unit 30. The integration unit 20 integrates the subject detection and tracking results corresponding to each of the multiple cameras. The inference unit 30 performs subject detection and tracking based on the images obtained from the cameras.

[0017] The integration unit 20 includes a multi-camera integration processing unit 21, a pipeline control determination unit 22, a pipeline output I / F 23, and a local information management unit 24.

[0018] The multi-camera integrated processing unit 21 includes a global tracking information group 212 and an integrated processing unit 213. Details of the integrated processing performed by the multi-camera integrated processing unit 21 will be explained later with reference to Figure 3.

[0019] The pipeline control determination unit 22 searches for a camera to acquire an image from when an event occurs, and instructs the camera to transmit the image. For example, the pipeline control determination unit 22 instructs a camera located within a certain distance from the location where the event occurred to transmit the image. The event may be, for example, a traffic accident such as a vehicle overturning or colliding on a road.

[0020] The pipeline control and determination unit 22 detects the occurrence of events and searches for cameras based on information acquired from APPs (Applications) 50 operating at various locations, including surveillance cameras and connected cars. The APPs 50 include road infrastructure APPs for accident and falling object prevention, urban development APPs for traffic flow analysis and prediction, autonomous driving support APPs, APPs installed in car navigation systems, and police-related APPs used to assist in accident response and verification.

[0021] The pipeline output interface 23 is an interface for outputting data from the integration unit 20 to the inference unit 30.

[0022] The local information management unit 24 includes a local tracking information group 241, an update unit 242, a coordinate transformation unit 243, a time mapping unit 244, and a pipeline input I / F 245.

[0023] The Local Tracking Information Group 241 is information obtainable from a single camera. For example, the Local Tracking Information Group 241 is information obtained when the camera detects an object (subject), and includes a Local Tracking ID (an ID that identifies the object), a Global Tracking ID (an ID that identifies the object), the time the object was detected, the coordinates of the location where the object was detected, the size of the object, the velocity of the object, and information for tracking the object (e.g., ReID, duration of stay). The Update Unit 242 updates the Local Tracking Information Group 241 (including saving new data).

[0024] The coordinate transformation unit 243 converts the coordinates of the object obtained by each camera into world coordinates. The time correspondence unit 244 associates the times when the object was detected by each camera with the times when the cameras were connected. In this way, the local information management unit 24 processes the information received from the inference unit 30 as appropriate and stores it as a local tracking information group 241.

[0025] The inference unit 30 includes a pipeline control unit 31, an estimation unit 32, an information acquisition unit 33, and an output interface 34. The pipeline control unit 31 controls the processing of the pipeline 321 by the estimation unit 32. The pipeline 321 is a pipeline corresponding to each of the multiple cameras.

[0026] The estimation unit 32 updates the Local tracking information group 322 in accordance with the processing of pipeline 321. The Local tracking information group 322 includes the camera ID, the time the object was detected, the result of the AI ​​that detected the object, the size of the object, the velocity of the object, information for tracking the object (e.g., ReID, time spent), and camera parameters.

[0027] The information acquisition unit 33 acquires information from the estimation unit 32. The output interface 34 outputs the information acquired by the information acquisition unit 33 to the integration unit 20.

[0028] Figure 2 is a flowchart showing the pipeline processing and integration processing flow according to the first embodiment.

[0029] The pipeline 321 includes pipelines 321_1, ..., and 321_n, each corresponding to camera 1, ..., camera n (where n is an integer greater than or equal to 2). Here, the processing of pipeline 321_1 will be described. However, the estimation unit 32 performs the same processing for all pipelines as for pipeline 321_1.

[0030] Pipeline 321_1 acquires images captured by camera 1 and reads information about the object being tracked by camera 1 from the Local Tracking Information Group 322. Pipeline 321_1 acquires multiple images and stores the acquired images in an image buffer. The images may be individual frames of a video.

[0031] Pipeline 321_1 repeats the processes from S102 to S109 for each image in the image buffer (steps S101, S110). Pipeline 321_1 receives an image input from the image buffer (step S102). Pipeline 321_1 detects the input image, i.e., the object in the current frame (step S103). Multiple objects may be detected at this time.

[0032] Pipeline 321_1 repeats steps S105 to S107 (steps S104, S108) for the number of objects detected (objects being tracked).

[0033] Pipeline 321_1 determines whether the detected object is a match, that is, whether the detected object is already stored in the Local tracking information group 233 (step S105). If there is no match (step S105; No), pipeline 321_1 deletes the object's information from the Local tracking information group 233 if it remains unmatched for a certain period of time (step S107). If there is a match (step S105; Yes), pipeline 321_1 updates the object's location and characteristic information in the Local tracking information group 233 (step S106).

[0034] Furthermore, if pipeline 321_1 finishes repeating in S108, it adds the unmatched detected object to the tracking target (step S109).

[0035] The information regarding the detected object obtained here is used in the integrated processing by the integrated processing unit 213.

[0036] The integrated processing unit 213 repeats the process from step S202 to S215 in increments of the DT (DateTime) update rate (steps S201, S216).

[0037] The integrated processing unit 213 extracts the target to be tracked at the current time from the Local tracking information group 241 (step S202).

[0038] The integrated processing unit 213 repeats steps S204 to S206 for each object being tracked in the Global tracking information group 212 (steps S203, S208). The integrated processing unit 213 checks whether the camera ID and Local tracking ID of the object being tracked globally (the object being tracked at the current time) are present in the Local tracking information group 241 (step S204).

[0039] If the object is not registered with an ID (step S205; No), the integrated processing unit 213 deletes the object being globally tracked from the global tracking information group 212 if it has not been updated for a certain period of time (step S207). If the object is registered with an ID (step S205; Yes), the integrated processing unit 213 updates the location and characteristics of the object in the global tracking information group 212 with the contents of the local tracking information group 241 (step S206).

[0040] Next, the integrated processing unit 213 repeats steps S210 to S213 (steps S209, S214) for each tracking target in the Local tracking information group 241 whose ID is not registered in the Global tracking information group 212 at the current time.

[0041] The integrated processing unit 213 matches the characteristics of the tracked object with the characteristics of a different camera ID in the Global Tracking Information Group 212 (step S210). If there is no match (step S211; No), the integrated processing unit 213 adds the ID of the tracked object to the Global Tracking Information Group 212 (step S213). If there is a match (step S211; Yes), the integrated processing unit 213 adds the camera ID and object information of the tracked object to the Global Tracking Information Group 212 (step S212).

[0042] When the repeat operation ends in S214, the integrated processing unit 213 selects a representative camera ID and a Global tracking ID (an ID that identifies an object) from the Global tracking information group 212 (step S215). The representative camera ID and Global tracking ID correspond to the representative tracking result. In other words, the integrated processing unit 213 switches the representative tracking result in step S215.

[0043] Figure 3 is a diagram illustrating the data flow related to the integrated processing. As shown in Figure 3, the estimation unit 32 performs tracking for each pipeline, i.e., event detection, position estimation, and velocity estimation (step S10).

[0044] Each time the integration processing unit 213 extracts a tracking target at the current time (time t) from the tracking results by the estimation unit 32, that is, the Global Tracking Information Group 212, it executes integration processing (steps S21, S22).

[0045] The matching by the integration processing unit 213 and the selection of the representative tracking result will be described using FIG. 4 and the like. FIG. 4 is a diagram for explaining the outline of the integration processing. In the example of FIG. 4, the processing when the integration processing unit 213 selects a representative tracking result from the tracking results of the subject 41, which is a person, obtained from the fixed camera (1), the moving camera (1), and the moving camera (2) will be described. Note that the fixed camera (1) is the surveillance camera 51. The moving camera (1) and the moving camera (2) are respectively mounted on the vehicle 52 and the vehicle 53.

[0046] The fixed camera (1), the moving camera (1), and the moving camera (2) each perform tracking of the subject 41. The tracking by a single camera may be performed by a known method.

[0047] The integration processing unit 213 collates the features obtained from the images between the cameras for each time window (time zone, time interval), and links the trajectories of different cameras. Whether there is a frame that is easy to collate or a frame that is difficult to collate depends on the time window. Therefore, if a subject is matched between a plurality of cameras at least once during a certain period, the integration processing unit 213 regards the subject as being the same. Further, the integration processing unit 213 may cross-compare the features of the image of a certain camera in the current time window with the features of the past time window of other cameras by looking at the history of the features of the image. Further, the integration processing unit 213 may perform matching by comparing not only the features of the image but also the position obtained by tracking, the distance of the trajectory, etc. together.

[0048] When attempting to track one desired object using multiple cameras, the object is not continuously imaged by all cameras. In the first time window, the object may be imaged only by some of the multiple cameras, such as the first camera and the second camera, and in the second time window, the first camera and the third camera. If the multiple cameras that image the object include a moving camera, it is also assumed that the object may be imaged only in the images of a few seconds within the time window imaged by the moving camera. It is conceivable to exclude the images captured by the camera that captures the object only in a very small number of time windows, but if the object is imaged at a different angle of view from other cameras, that is, if the amount of information in a very small number of time windows is rich compared to other cameras. Therefore, the integration processing unit 213 may be designed to regard as the target any one that has been matched with other cameras for each time window.

[0049] The integration processing unit 213 selects the tracking result of one camera as the representative tracking result from the images in which the object is matched for each time window. Also, the integration processing unit 213 may hold the tracking results that were not selected without discarding them.

[0050] In the example of FIG. 4, the tracking result of the fixed camera (1) is selected as the representative tracking result and reflected in the digital twin at time t0 + Δt.

[0051] In this way, the integration processing unit 213 is a set including at least a part of the multiple cameras, and for each camera included in the set in which the components change due to the movement of each of the multiple cameras, it acquires the tracking result which is the result of detecting and tracking the subject. The integration processing unit 213 selects one tracking result that meets the criteria as the representative tracking result from the tracking results.

[0052] The fixed camera (1), the moving camera (1), and the moving camera (2) are an example of the set. For example, since the position of the moving camera (1) changes according to the movement of the vehicle 52, it may fall outside the range where the subject can be tracked and be excluded from the set.

[0053] The integrated processing unit 213 may select as the representative tracking result the one in which the subject 41 was first detected within the same time window, the one in which the size of the subject 41 is largest based on the image size, or the one in which the distance between the camera and the subject is closest. Alternatively, the integrated processing unit 213 may select as the representative tracking result the one in which tracking of the subject 41 has been successfully performed continuously for the longest period within the same time window. Furthermore, the integrated processing unit 213 may select as the representative tracking result an image taken with a camera that has been selected as the representative tracking result in a past or future time window. Additionally, the integrated processing unit 213 may select as the representative tracking result an image taken with a camera that has been successfully tracking continuously in a past time window. Furthermore, if there are multiple candidates, the integrated processing unit 213 may select as the representative tracking result an image taken with a camera that has been successfully tracking continuously from the longest period (oldest time window).

[0054] Thus, the integrated processing unit 213 can select a tracking result as a representative tracking result that satisfies criteria based on at least one of the following: the confidence level set for each of the multiple cameras, the order in which each of the multiple cameras detected the subject for each time window, the confidence level of each of the multiple cameras in detecting the subject, the distance between each of the multiple cameras and the subject, the performance of each of the multiple cameras, and representative tracking results selected in the past. Furthermore, if the performance and quality of the cameras are known, the integrated processing unit 213 may prioritize and adopt the tracking result obtained by the camera with higher specific parameters, such as image resolution and frame rate.

[0055] The integrated processing unit 213 does not need to select a representative tracking result for every frame; it may select a new representative tracking result if the subject 41 is lost in a previously selected representative tracking result. Furthermore, the time window may switch if the subject 41 is lost in a previously selected representative tracking result. Also, the representative tracking results may later be arbitrarily selected and edited by the user.

[0056] Thus, the integrated processing unit 213 focuses on the fact that tracking individual cameras is easier than matching cameras, and manages the tracking results of individual cameras independently, using camera matching only to link the tracking results of cameras together.

[0057] This prevents misidentification of subjects, as well as overlooking or replacing subjects. As a result, this embodiment can reduce the discrepancy between the tracking results and the facts occurring in real space.

[0058] Furthermore, the integrated processing unit 213 does not discard the tracking results from each camera, but can select a representative tracking result for each time period and filter out tracking results other than the representative one.

[0059] Figure 5 illustrates an example of the integrated processing. In the example in Figure 5, tracking of a subject 41, which is a person, is performed, that is, the estimation of the true person's trajectory. The tracking is performed by a fixed camera (1), a mobile camera (1), a mobile camera (2), a mobile camera (3), and a mobile camera (4). The fixed camera (1) is a surveillance camera 55. The mobile cameras (1), (2), (3), and (4) are mounted on vehicles 56, 57, 58, and 59, respectively.

[0060] Each camera's tracking results are assigned a Local Tracking ID. For example, the Local Tracking ID for subject 41 in the tracking of fixed camera (1) is "x". Also, for example, the Local Tracking ID for subject 41 in the tracking of moving camera (2) is "z". Thus, even if the subject is the same, the Local Tracking IDs are different.

[0061] The integration processing unit 213 calculates similarity in image features, position, velocity, etc., in the background in order to link tracking results that have different Local Tracking IDs. For example, if the integration processing unit 213 finds a subject for the first time in any of the tracking results, it assigns a Global Tracking ID to that subject. Then, the integration processing unit 213 integrates the other tracking results with that Global Tracking ID.

[0062] For example, when the subject 41 is first found in the tracking results of the fixed camera (1), the integrated processing unit 213 issues a Global tracking ID "n" and associates the Local tracking ID "x" of the fixed camera (1) with the Global tracking ID "n". Subsequently, when the subject 41 is found in the tracking results of other cameras, the Local tracking IDs "y", "z", "u", and "w" are further associated with the Global tracking ID "n".

[0063] It is also possible that the integrated processing unit 213 may mistakenly identify subject 41 as a different subject and issue a different Global Tracking ID "m". In such a case, if the subject associated with Global Tracking ID "m" later matches the subject associated with Global Tracking ID "n", Global Tracking ID "m" may be discarded, and the Local Tracking ID that was associated with Global Tracking ID "m" may be associated with Global Tracking ID "n".

[0064] Furthermore, the integrated processing unit 213 may provide a UI that allows the user to change or delete the mapping between the Global Tracking ID and the Local Tracking ID if the user visually confirms that the mapping is incorrect.

[0065] The tracking results shown in Figure 5 are obtained by connecting the representative tracking results selected by the integrated processing unit 213. A switch indicates that the representative tracking result has changed. For example, the correspondence between the time window and the selected representative tracking result is as follows: Time window 00:00 to 00:03: Moving camera (1) Time window 00:03 to 00:05: Moving camera (3) Time window 00:05 to 00:10: Moving camera (2) Time window 00:10 to 00:18: Fixed camera (1) Time window 00:18 to 00:23: Moving camera (4)

[0066] For example, the subject's trajectory up to point 61 is the tracking result of the mobile camera (1). The subject's trajectory from point 61 to point 62 is the tracking result of the mobile camera (3). The subject's trajectory from point 62 to point 63 is the tracking result of the mobile camera (2). The subject's trajectory from point 63 to point 64 is the tracking result of the fixed camera (1). The trajectory from point 64 onward is the tracking result of the mobile camera (4).

[0067] Furthermore, if tracking results from highly reliable cameras, such as stable fixed cameras or cameras close to the event, exist within the group, the integrated processing unit 213 may select the tracking results from such highly reliable cameras as the representative tracking results.

[0068] Furthermore, when the representative tracking result is switched, if there is a large difference in position before and after the switch, the integrated processing unit 213 may, for the purpose of mitigating that difference, take an average of the representative tracking results before and after the switch for a certain period after the switch, or perform correction using a Kalman filter or the like.

[0069] The following shows examples of rules and determination methods for when the integrated processing unit 213 selects a representative tracking result from the tracking results of multiple cameras. For example, the integrated processing unit 213 refers to the rules from top to bottom and selects the first tracking result that satisfies the rule. If there are multiple tracking results that satisfy a certain rule, the integrated processing unit 213 refers to further lower-level rules.

[0070] (Rule 1) Tracking is still ongoing. (Determination method) In the tracking results of the camera alone, the subject's trajectory has not been invalidated (for example, trajectory invalidation occurs when the subject is not matched continuously for a certain period of time or longer within the camera. In other words, the condition is to select a tracking result in which the number of consecutive time intervals in which the subject was not matched is below or less than a predetermined threshold).

[0071] For example, during tracking, there may be instances where the system fails to match some frames or subjects. Therefore, the integrated processing unit 213 may determine that the tracking result with the oldest initial matching time is the tracking result with the longest continuous tracking period. Since continuous tracking is generally difficult, even if tracking is interrupted, if tracking can be restarted within a predetermined time, it is considered to have continued. However, since it is possible to track a different subject when tracking is restarted, this rule may be combined with other rules.

[0072] (Rule 2) The camera has been tracking the subject for the longest time. (Determination method) In the tracking results of the camera alone, the camera with the oldest tracking start time is selected.

[0073] (Rule 3) The most stable trajectory. (Determination method) The error is small compared to the result of smoothing each camera's trajectory with a Kalman filter. For example, the tracking result corresponding to the trajectory with coordinates closest to the coordinates of the current time predicted by applying a Kalman filter to the trajectory of the integrated tracking result is selected. A stable trajectory can also be rephrased as having little fluctuation in movement speed (small difference between maximum and minimum speed), not being excessively zigzag (the movement trajectory is close to a straight line), etc.

[0074] (Rule 4) At the most recent time, the camera is not undetected (hidden, missed). (Determination method) Based on the tracking results of the camera alone, the matching target is not lost at the most recent time.

[0075] (Rule 5) The camera is close to the subject. (Determination method) The distance between the camera and the subject in world coordinates is small (within a certain value), or the subject appears large in the camera image (above a certain size).

[0076] (Rule 6) The detected object is not cut off or distorted. (Determination method) The bounding rectangle of the object detection result at the current time does not touch the edge of the image. There is no similar bounding rectangle in front of or in front of the bounding rectangle. Apply another AI model to determine whether or not the object is cut off.

[0077] (Rule 7) The object to be detected has high visibility. (Determination method) The confidence level at the time of object detection is above a certain level.

[0078] (Rule 8) The camera being used has high hardware reliability. (Determination method) It is a fixed camera, not a moving camera. It is a camera with high resolution and bitrate. The camera is mounted high (field of view with minimal overlap). It is a model recommended by the system (especially in cases where there is a large variation in quality, such as in-car cameras in connected cars).

[0079] [Program] In one embodiment, the tracking device (server 10) can be implemented by installing a tracking program that performs the above processing as packaged software or online software on a desired computer. For example, by having the above tracking program run on an information processing device, the information processing device can be made to function as a tracking device. The information processing device referred to here includes desktop or notebook personal computers. In addition, the information processing device also includes smartphones, tablet terminals, etc.

[0080] Figure 6 shows an example of a computer running a tracking program. Computer 1000 has, for example, memory 1010 and a CPU 1020. Computer 1000 also has a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0081] Memory 1010 includes ROM (Read Only Memory) 1011 and RAM (Random Access Memory) 1012. ROM 1011 stores, for example, a boot program such as BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to the hard disk drive 1090. The disk drive interface 1040 is connected to the disk drive 1100. For example, a removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to, for example, a mouse 1110 and a keyboard 1120. The video adapter 1060 is connected to, for example, a display 1130.

[0082] The hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. That is, the program that defines each process of the tracking device is implemented as a program module 1093 in which executable code for a computer is written. The program module 1093 is stored, for example, in the hard disk drive 1090. For example, a program module 1093 for performing the same processes as the functional configuration in the tracking device is stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced by an SSD.

[0083] Furthermore, the configuration data used in the processing of the above-described embodiment is stored as program data 1094 in, for example, memory 1010 or hard disk drive 1090. The CPU 1020 then reads the program module 1093 and program data 1094 stored in memory 1010 or hard disk drive 1090 into RAM 1012 as needed and executes the processing of the above-described embodiment.

[0084] Furthermore, the program module 1093 and program data 1094 are not limited to being stored in the hard disk drive 1090; for example, they may be stored in a removable storage medium and read by the CPU 1020 via a disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (LAN (Local Area Network), WAN (Wide Area Network), etc.). The program module 1093 and program data 1094 may then be read by the CPU 1020 from the other computer via a network interface 1070.

[0085] Although embodiments applying the invention made by the present inventors have been described above, the present invention is not limited by the descriptions and drawings that constitute part of the disclosure of the present invention in this embodiment. That is, all other embodiments, examples, and operational techniques made by those skilled in the art based on this embodiment are included in the scope of the present invention.

[0086] 10 Server 20 Integration Unit 21 Multi-camera Integration Processing Unit 22 Pipeline Control Judgment Unit 23 Output I / F for Pipeline 24 Local Information Management Unit 30 Inference Unit 31 Pipeline Control Unit 32 Estimation Unit 33 Information Acquisition Unit 34 Output I / F 41 Subject 50 APP 51, 55 Surveillance Camera 52, 53, 56, 57, 58, 59 Vehicle 61, 62, 63, 64 Location 212 Global Tracking Information Group 213 Integration Processing Unit 241, 322 Local Tracking Information Group 242 Update Unit 243 Coordinate Transformation Unit 244 Time Correspondence Unit 245 Input I / F for Pipeline 321, 321_1, 321_n Pipeline

Claims

1. A tracking method characterized in that a computer performs a process of obtaining tracking results, which are the results of each camera in a set comprising at least some of a plurality of cameras as constituent elements, wherein the constituent elements change as each of the plurality of cameras moves, and selecting one tracking result from the aforementioned tracking results that satisfies a criterion as a representative tracking result.

2. The tracking method according to claim 1, wherein the tracking result is time-series information, and the selecting process selects a representative tracking result from the tracking results for each time window that switches at predetermined timings.

3. The tracking method according to claim 2, characterized in that the selection process selects as the representative tracking result a tracking result that satisfies a criterion based on at least one of the following: the confidence level set for each of the plurality of cameras, the order in which each of the plurality of cameras detected the subject for each time window, the confidence level of each of the plurality of cameras in detecting the subject, the distance between each of the plurality of cameras and the subject, the performance of each of the plurality of cameras, and the representative tracking result selected in the past.

4. The tracking method according to claim 3, characterized in that the time window is switched when the camera corresponding to the representative tracking result loses the subject.

5. A tracking device characterized by acquiring tracking results, which are the results of each camera in a set comprising at least some of a plurality of cameras as components, wherein the components change as each of the plurality of cameras moves, and selecting one tracking result that satisfies a criterion from among the tracking results as a representative tracking result.

6. A tracking program characterized by causing a computer to perform a process of obtaining tracking results, which are the results of each camera in a set comprising at least some of a plurality of cameras as constituent elements, wherein the constituent elements change as each of the plurality of cameras moves, and selecting one tracking result that satisfies a criterion from among the tracking results as a representative tracking result.

Images