Object tracking device, object tracking method, and program

Abstract

An information processing device 100 according to the present disclosure comprises: a segmentation unit 121 that extracts, from a time-series image, a plurality of object candidates for one object corresponding to designated object information; and a tracking unit 122 that identifies and tracks an object to be tracked from the plurality of object candidates on the basis of a tracking result of the object for a past image.

Description

Object tracking device, object tracking method, and program

[0001] The present disclosure relates to an object tracking device, an object tracking method, and a program.

[0002] As described in Patent Document 1, an object is tracked from an image sequence. For example, in Patent Document 1, an object is tracked in an image frame using a template. In addition, a technique called SAM (Segment Anything Model) for extracting an object from an image has become known in recent years.

[0003] Japanese Patent Application Laid-Open No. 2001-273501

[0004] However, when tracking an object from a sequence of images using the above-mentioned techniques, a different object may be tracked due to changes in the images, resulting in a problem of reduced tracking accuracy.

[0005] Therefore, an object of the present disclosure is to solve the above-mentioned problem of reduced tracking accuracy when tracking an object from an image.

[0006] An object tracking device according to one embodiment of the present disclosure includes: a segmentation unit that extracts, from time-series images, a plurality of object candidates for one object corresponding to specified object information; and a tracking unit that identifies a tracking target from the plurality of object candidates based on object tracking results for past images and tracks the tracking target. An object tracking method according to one embodiment of the present disclosure includes: extracting, from time-series images, a plurality of object candidates for one object corresponding to specified object information; identifying a tracking target from the plurality of object candidates based on object tracking results for past images and tracking the tracking target. A program according to one embodiment of the present disclosure includes: causing a computer to execute processing to extract, from time-series images, a plurality of object candidates for one object corresponding to specified object information; and identifying a tracking target from the plurality of object candidates based on object tracking results for past images and tracking the tracking target.

[0007] With the above-described configuration, the present disclosure can suppress a decrease in tracking accuracy when tracking an object from an image.

[0008] FIG. 1 is a block diagram showing the configuration of an object tracking device according to the present disclosure. FIG. 2 is a diagram showing the processing state of an object tracking device according to the present disclosure. FIG. 3 is a diagram showing the processing state of an object tracking device according to the present disclosure. FIG. 4 is a flowchart showing the processing operation of an object tracking device according to the present disclosure. FIG. 5 is a block diagram showing the hardware configuration of an object tracking device according to the present disclosure. FIG. 6 is a block diagram showing the configuration of an object tracking device according to the present disclosure.

[0009] First Embodiment A first embodiment of the present disclosure will be described with reference to the drawings. Note that the drawings may be relevant to any embodiment.

[0010] The object tracking device 10 in this embodiment is used to track objects captured in a video consisting of a time-series of image frames (image sequence). At this time, the object tracking device 10 extracts an object from each image frame to identify the tracking target, but this embodiment is particularly characterized in that it extracts multiple object candidates corresponding to one object from the image frames and identifies the tracking target from these multiple object candidates.

[0011] The object tracking device 10 is composed of one or more information processing devices each including a calculation device and a storage device. As shown in FIG. 1 , the object tracking device 10 includes a target designation unit 11, a segmentation unit 12, a tracking unit 13, and a prediction unit 14. The functions of the target designation unit 11, the segmentation unit 12, the tracking unit 13, and the prediction unit 14 can be realized by the calculation device executing a program for realizing each function stored in the storage device. The object tracking device 10 is also configured to store target information 15, segmentation information 16, tracking information 17, and prediction information 18 in the storage device. The functions and operations of each component will be described below.

[0012] The object tracking device 10 configured as described above first receives input of video G consisting of a time-series of image frames (image sequence). The target designation unit 11 of the object tracking device 10 also receives input of target information, which is information about the object to be tracked, from the operator (step S1 in FIG. 5). For example, the target information may be frame information surrounding the object to be tracked, and the image information within this frame information corresponds to the object that is the target of tracking. The case where the target information is frame information will be described below as pattern A.

[0013] Specifically, when receiving frame information as target information in pattern A, the target designation unit 11 outputs a predetermined image frame of the input video G to the operator's information processing terminal for display, and receives input of frame information representing a predetermined area within the image frame designated by the operator for the image frame displayed on the information processing terminal. The frame information may be information designated by the operator performing an operation such as surrounding a predetermined area, or may be information designated by the operator performing an operation such as selecting an object shown in the image frame and automatically setting an area surrounding the selected object using the function of the target designation unit 11. Furthermore, the target designation unit 11 may receive input of frame information representing a predetermined area within an image by receiving information in which frame information is attached to a predetermined image from the operator's information processing terminal. In this case, the frame information may be graphic information surrounded by a rectangle or curve, or may be text information such as coordinate information specifying the area.

[0014] The target designation unit 11 may also accept text information as another example of target information. The case where the target information is text information will be described below as pattern B. Specifically, the target designation unit 11 accepts text information describing an object, which is composed of text data input from the operator's information processing terminal. The text information is a word or a sentence formed by connecting words, and examples include "suspicious person," "person with a knife," "loitering person," and "emergency vehicle." In this case, when the target designation unit 11 accepts input of text information composed of an abstract sentence such as "suspicious person," it may interpret the "suspicious person" into a specific object and accept the resulting text information as target information. For example, it may interpret the abstract text information of "suspicious person" into a specific object such as "person with a knife" or "loitering person," and accept the interpreted text information as target information. The target designation unit 11 may interpret the input abstract text information into specific text information using any method, such as using a large language model (LLM).

[0015] The segmentation unit 12 sequentially extracts object candidates corresponding to the above-mentioned target information from each image frame constituting the input video in chronological order. In this embodiment, the segmentation unit 12 extracts multiple object candidates for one object corresponding to the target information in one image frame (step S2 in FIG. 5). For example, the segmentation unit 12 extracts, as the object candidate information, information representing an area in the image frame in which the object candidate corresponding to the target information appears.

[0016] Here, when extracting object candidates from an image frame, the segmentation unit 12 performs the extraction process by setting parameter values ​​used in the extraction process. For example, the segmentation unit 12 extracts object candidates by setting parameter values ​​that may affect the degree, reliability, and accuracy of object candidate extraction, such as the size value of the object candidate or the number of feature points of the object candidate. At this time, the segmentation unit 12 perturbs the parameter values ​​to generate multiple parameter values. As an example, if a reference value for the parameter value is set in advance, the reference value is perturbed, i.e., changed from the reference value, to generate multiple parameter values ​​consisting of the reference value and the change value. Then, the segmentation unit 12 performs the object candidate extraction process by setting the generated parameter values, respectively, and extracts multiple object candidates corresponding to the respective parameters. For example, if the parameter is the size value of the object candidate described above, the reference value is a preset value or the size value of the object when it was identified as the tracked target in the previous image frame, and the change value is a value that is increased or decreased by a preset percentage from the reference value. The parameters are not limited to the size and the number of feature points of the object candidates described above, but may be any type of parameters that are set when extracting object candidates from an image frame.

[0017] Specifically, in this embodiment, the segmentation unit 12 extracts multiple object candidates corresponding to the target information using a technique for extracting objects from an image called SAM (Segment Anything Model). At this time, the segmentation unit 12 generates multiple parameter values ​​by perturbing the values ​​of parameters such as "pred_iou_thresh" and "points_per_side," which are SAM parameters that set the size and number of feature points of the object to be extracted, and extracts object candidates corresponding to the target information from the image frame t with the respective parameter values ​​set.

[0018] 2 and 3, an example of the process performed by the segmentation unit 12 to extract object candidates corresponding to target information from an image will be described. In this example, a case will be described in which the t-th image frame as shown in FIG. 3 is processed from a sequence of images constituting video G. In this example, the target information is pattern A, which is frame information, and for example, the image information to be targeted for tracking within the frame information represents the characteristics of person P as shown in image t shown on the left side of FIG. 3. Furthermore, in this example, the value of the size of the object region to be extracted is a parameter used during the extraction process.

[0019] First, the segmentation unit 12 sets a parameter that is a value of the size of the extracted region in the SAM to a reference value, and segments object candidates that represent the region of person P corresponding to the target information from the image frame t (reference numeral 12a in FIG. 2). As a result, an object region such as that indicated by reference numeral R1 is extracted, as shown in the image frame in the middle right section of FIG. 3, and information about this object region is stored as segmentation information 16. Note that while the example in FIG. 3 illustrates a case where object region R1 is elliptical, it may be a region of any shape, such as a region surrounded by complex curves.

[0020] Next, the segmentation unit 12 changes and sets the parameter for the size of the extraction region in the SAM to a value larger than the reference value (reference numeral 12b in FIG. 2 ), and segments an object candidate representing the region of the person P corresponding to the target information from the image frame t (reference numeral 12c in FIG. 2 ). As a result, an object region such as that indicated by reference numeral R2 is extracted, as shown in the image frame in the upper right section of FIG. 3 , and information about this object region is stored as segmentation information 16. In the example of FIG. 3 , because the extraction region is large, an object candidate R2 from a region that also includes a background image may be extracted. However, if the person P appears large in the image frame, an object candidate R2 appropriate for the size of the person P may be extracted.

[0021] Next, the segmentation unit 12 changes and sets the parameter for the size of the extraction region in the SAM to a value smaller than the reference value (reference numeral 12d in FIG. 2 ), and segments an object candidate representing the region of the person P corresponding to the target information from the image frame t (reference numeral 12e in FIG. 2 ). As a result, an object region such as that indicated by reference numeral R3 is extracted, as shown in the image frame in the lower right section of FIG. 3 , and information about this object region is stored as segmentation information 16. In the example of FIG. 3 , because the extraction region is small, an object candidate R3 representing only a portion of the region of the person P may be extracted. However, if the person P appears small in the image frame, an object candidate R3 appropriate for the size of the person P may be extracted.

[0022] In this manner, the segmentation unit 12 extracts three object candidates R1, R2, and R3 for one person P from the image frame t, as shown in FIG. 3 . Note that if more parameter values ​​are generated, even more object candidates will be extracted. If additional target information is present in the image frame t, the segmentation unit 12 extracts the object candidates described above for each piece of target information. That is, the segmentation unit 12 repeatedly performs the process surrounded by reference numeral 12A in FIG. 2 for each piece of target information. Note that, even when the target information is text information, pattern B, multiple object regions are extracted for one image frame in the same manner as described above. That is, when the target information is text information such as "a person wandering around" and corresponds to person P shown in FIG. 3 , object regions R1, R2, and R3, which are regions of multiple people, are extracted as shown on the right side of FIG. 3 .

[0023] Furthermore, the segmentation unit 12 may vary and set a plurality of parameters in the SAM, and extract object candidates for each parameter setting. For example, when varying and setting the size and number of feature points of the object candidates as parameters, combinations are set in which the size is varied from a reference value to a large value and a small value, and the number of feature points is varied from a reference value to a large value and a small value, and object candidates are extracted for each combination of parameter settings.

[0024] The tracking unit 13 identifies a tracking target from a plurality of object candidates extracted from image frame t as described above (step S3 in FIG. 5 ) and tracks the identified tracking target (step S4 in FIG. 5 ). At this time, the tracking unit 13 identifies the tracking target from a plurality of object candidates based on object tracking results for past image frames. Specifically, the tracking unit 13 performs tracking prediction of an object in the t-th image frame currently being processed based on object tracking results for image frames 1 to t-1 prior to the t-th image frame, and identifies the tracking target from a plurality of object candidates based on this tracking prediction. In this way, since tracking prediction is used to identify the tracking target, the prediction unit 14 predicts the position of the tracking target in a subsequent image frame each time tracking is performed in a previous image frame by the tracking unit 13. For example, the prediction unit 14 performs tracking prediction of the tracking target in a subsequent image frame using an existing prediction method (e.g., a Kalman filter) based on tracking results for previous image frames for each tracking target, and stores the prediction information 18. As an example, the prediction unit 14 may use the position or area of ​​the tracking target identified in the most recent image frame t-1 as the tracking prediction of the tracking target in the next image frame t.

[0025] 4, the process of identifying a tracking target from a plurality of object candidates in the tracking process by the tracking unit 13 will be described in detail. Note that the following description will be divided into two cases: pattern A in which the target information is frame information, and pattern B in which the target information is text information. Also, it is assumed that a predicted region of the tracking target in a subsequent image t is stored as prediction information 18 as a tracking prediction made by the prediction unit 14 from the tracking results of past image frames 1 to (t-1).

[0026] First, the tracking unit 13 compares the predicted region of the tracking target in image t with the regions of multiple object candidates to check the degree of overlap (IoU: Intersection over Union) between them (reference numeral 13a in FIG. 4). Then, the tracking unit 13 selects the object candidate that overlaps the most with the predicted region as the object candidate associated with the predicted region, identifies this object candidate as the tracking target, and registers it in the tracking information 17 (reference numeral 13b in FIG. 4). Note that other object candidates that are not associated with the predicted region are discarded (reference numeral 13c in FIG. 4).

[0027] On the other hand, as described above, when the tracking unit 13 determines that the degree of overlap between the predicted region and each of the regions of the multiple object candidates is low and that none of the object candidates are associated with the predicted region, it determines that the image frame t is an initial frame or that the object candidate is a newly appeared object (reference numeral 13d in FIG. 4). In other words, in this case, it determines that the tracking target cannot be identified from the tracking prediction, and identifies and registers a new tracking target from the multiple object candidates (reference numeral 13e in FIG. 4).

[0028] Here, in the case of pattern A where the target information is frame information, it is determined whether the image frame t is a new frame, and if it is a new frame, the object candidate extracted by the segmentation process using the reference value is specified and registered as a new tracking target. On the other hand, if the image frame t is not a new frame, the frame information which is the target information is referenced, and the object candidate whose area overlaps the area of ​​the frame information most with that of each of the multiple object candidates is registered as a new tracking target.

[0029] Furthermore, in the case of pattern B in which the target information is text information, a word string is generated from each of the multiple object candidates, regardless of whether the image frame t is a new frame or not. For example, an existing LLM or the like is used to generate a word string from the image of each object candidate. Then, by referring to the text information that is the target information, a certain object candidate is registered as a new tracking target depending on the degree of overlap between the text information and the word strings of each of the multiple object candidates. For example, the object candidate with the highest matching rate or distance between the characters contained in the text information and the word string of the object candidate is registered as a new tracking target. This allows tracking of the new tracking target to be performed in the same manner as described above.

[0030] In addition, by the above-mentioned processing, tracking predictions that are not associated with object candidates are updated as tracking predictions for the subsequent image frame t+1, and if such updates occur repeatedly, the registration of the tracking target associated with the tracking prediction is deleted.

[0031] As described above, in this embodiment, multiple object candidates corresponding to one object are extracted from the image frame, and a tracking target is identified and tracked from the multiple object candidates, thereby preventing a decrease in object tracking accuracy.

[0032] Second Embodiment Next, a second embodiment of the present disclosure will be described with reference to the drawings. This embodiment shows an outline of the configuration of the object tracking device described in the above embodiment. Note that Figures 6 and 7 are diagrams for explaining the configuration, and these diagrams may be relevant to any of the embodiments.

[0033] First, the hardware configuration of the object tracking device 100 will be described with reference to Fig. 6. The object tracking device 100 is configured as a general information processing device, and is equipped with the following hardware configuration, for example: CPU (Central Processing Unit) 101 (arithmetic unit); ROM (Read Only Memory) 102 (storage device); RAM (Random Access Memory) 103 (storage device); programs 104 loaded into RAM 103; storage device 105 storing programs 104; drive device 106 for reading and writing data from and to a storage medium 110 external to the information processing device; communication interface 107 for connecting to a communication network 111 external to the information processing device; input / output interface 108 for inputting and outputting data; and bus 109 for connecting the various components.

[0034] 6 shows an example of the hardware configuration of the information processing device that is the object tracking device 100, and the hardware configuration of the information processing device is not limited to the above-described case. For example, the information processing device may be configured with a part of the above-described configuration, such as not including the drive device 106. Furthermore, instead of the above-described CPU, the information processing device may use a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point Number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof.

[0035] The object tracking device 100 can be equipped with the segmentation unit 121 and the tracking unit 122 shown in Fig. 7 by having the CPU 101 acquire and execute the program group 104. The program group 104 is stored in advance in the storage device 105 or the ROM 102, for example, and is loaded into the RAM 103 and executed by the CPU 101 as needed. The program group 104 may be supplied to the CPU 101 via the communication network 111, or may be stored in advance in the storage medium 110, and the drive device 106 may read out the program and supply it to the CPU 101. However, the segmentation unit 121 and the tracking unit 122 described above may be constructed using dedicated electronic circuits for realizing such means.

[0036] The segmentation unit 121 extracts a plurality of object candidates for one object corresponding to designated object information from time-series images. The tracking unit 122 identifies a tracking target from the plurality of object candidates based on object tracking results for past images, and tracks the tracking target.

[0037] With the above-described configuration, the present disclosure extracts multiple object candidates corresponding to one object from an image, and identifies and tracks a tracking target from the multiple object candidates, thereby suppressing a decrease in object tracking accuracy.

[0038] In addition, at least one of the functions of the segmentation unit 121 and the tracking unit 122 described above may be executed by an information processing device installed and connected anywhere on the network, that is, they may be executed by so-called cloud computing.

[0039] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-RWs, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program can also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can be supplied to a computer via wired communication paths such as electric wires and optical fibers, or via wireless communication paths.

[0040] Although the present disclosure has been described above with reference to the above-described embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each of the above-described embodiments can be combined with other embodiments as appropriate.

[0041] <Supplementary Notes> Some or all of the above embodiments can also be described as in the following supplementary notes. Below, an outline of the configurations of the object tracking device, object tracking method, and program according to the present disclosure will be described. However, the present disclosure is not limited to the following configurations. (Supplementary Note 1) An object tracking device comprising: a segmentation unit that extracts multiple object candidates for one object corresponding to specified object information from time-series images; and a tracking unit that identifies a tracking target from the multiple object candidates based on object tracking results for past images and tracks the tracking target. (Supplementary Note 2) The object tracking device according to Supplementary Note 1, wherein the segmentation unit generates multiple parameter values ​​that are set when performing processing to extract the object candidates, and extracts the multiple object candidates by setting the multiple parameter values ​​respectively. (Supplementary Note 3) The object tracking device according to Supplementary Note 2, wherein the segmentation unit generates multiple parameter values ​​that set a degree to which the object candidates are extracted, and extracts the multiple object candidates by setting the multiple parameter values ​​respectively. (Supplementary Note 4) The object tracking device according to Supplementary Note 2, wherein the segmentation unit generates a plurality of values ​​for the parameter that sets the size of the object candidates to be extracted, and extracts the plurality of object candidates by setting the plurality of parameter values, respectively. (Supplementary Note 5) The object tracking device according to Supplementary Note 2, wherein the segmentation unit generates a plurality of values ​​for the parameter that sets the number of feature points of the object candidates to be extracted, and extracts the plurality of object candidates by setting the plurality of parameter values, respectively. (Supplementary Note 6) The object tracking device according to Supplementary Note 1, wherein the tracking unit identifies the tracking target from the plurality of object candidates based on a tracking prediction for a future image that is based on object tracking results for past images. (Supplementary Note 7) The object tracking device according to Supplementary Note 6, wherein the tracking unit identifies the tracking target based on a comparison between the tracking prediction that represents a predicted object region in a future image that is based on object tracking results for past images and the region of each of the plurality of object candidates.(Supplementary Note 8) The object tracking device according to Supplementary Note 6, wherein the tracking unit, when unable to identify the tracking target based on the tracking prediction, identifies the tracking target based on a comparison between the specified object information and information based on the plurality of object candidates. (Supplementary Note 9) An object tracking method, comprising: extracting a plurality of object candidates for one object corresponding to specified object information from time-series images; identifying a tracking target from the plurality of object candidates based on object tracking results for past images, and tracking the tracking target. (Supplementary Note 10) The object tracking method according to Supplementary Note 9, wherein a plurality of parameter values ​​are generated to be set when performing processing to extract the object candidates, and the plurality of object candidates are extracted by respectively setting the plurality of parameter values. (Supplementary Note 11) The object tracking method according to Supplementary Note 10, wherein a plurality of parameter values ​​are generated to set a degree to which the object candidates are extracted, and the plurality of object candidates are extracted by respectively setting the plurality of parameter values. (Supplementary Note 11.1) The object tracking method according to Supplementary Note 10, wherein a plurality of values ​​for the parameter that sets the size of the object candidate to be extracted are generated, and the plurality of object candidates are extracted by setting the plurality of parameter values, respectively. (Supplementary Note 11.2) The object tracking method according to Supplementary Note 10, wherein a plurality of values ​​for the parameter that sets the number of feature points of the object candidate to be extracted are generated, and the plurality of object candidates are extracted by setting the plurality of parameter values, respectively. (Supplementary Note 12) The object tracking method according to Supplementary Note 9, wherein the tracked target is identified from the plurality of object candidates based on a tracking prediction for a future image that is based on object tracking results for past images. (Supplementary Note 13) The object tracking method according to Supplementary Note 12, wherein the tracked target is identified based on a comparison of the tracking prediction that represents a predicted object region in a future image that is based on object tracking results for past images with the region of each of the plurality of object candidates.(Supplementary Note 14) The object tracking method according to Supplementary Note 12, wherein, when the tracking target cannot be identified based on the tracking prediction, the tracking target is identified based on a comparison of the specified object information with information based on a plurality of the object candidates. (Supplementary Note 15) A computer-readable storage medium storing a program causing a computer to execute processes of: extracting, from time-series images, a plurality of object candidates for one object corresponding to specified object information, identifying a tracking target from the plurality of object candidates based on object tracking results for past images, and tracking the tracking target.

[0042] REFERENCE SIGNS LIST 10 Object tracking device 11 Target designation unit 12 Segmentation unit 13 Tracking unit 14 Prediction unit 15 Target information 16 Segmentation information 17 Tracking information 18 Prediction information 100 Object tracking device 101 CPU 102 ROM 103 RAM 104 Program group 105 Storage device 106 Drive device 107 Communication interface 108 Input / output interface 109 Bus 110 Storage medium 111 Communication network 121 Segmentation unit 122 Tracking unit

Claims

1. An object tracking device comprising: a segmentation unit that extracts multiple object candidates for one object corresponding to specified object information from a time series of images; and a tracking unit that identifies a tracking target from the multiple object candidates based on object tracking results for past images and tracks the tracking target.

2. An object tracking device according to claim 1, wherein the segmentation unit generates a plurality of parameter values ​​to be set when performing the process of extracting the object candidates, and extracts a plurality of the object candidates by setting the values ​​of the plurality of parameters, respectively.

3. An object tracking device according to claim 2, wherein the segmentation unit generates a plurality of values ​​for the parameter that sets the degree to which the object candidates are extracted, and extracts a plurality of the object candidates by setting the values ​​for the plurality of parameters, respectively.

4. An object tracking device according to claim 2, wherein the segmentation unit generates a plurality of values ​​for the parameter that sets the size of the object candidate to be extracted, and extracts a plurality of the object candidates by setting the values ​​for the plurality of parameters, respectively.

5. An object tracking device according to claim 2, wherein the segmentation unit generates a plurality of values ​​for the parameter that sets the number of feature points of the object candidate to be extracted, and extracts a plurality of the object candidates by setting the values ​​for the plurality of parameters, respectively.

6. An object tracking device according to claim 1, wherein the tracking unit identifies the tracking target from a plurality of object candidates based on tracking prediction for a future image based on object tracking results for past images.

7. An object tracking device according to claim 6, wherein the tracking unit identifies the tracking target based on a comparison between the tracking prediction, which represents a predicted object area in a subsequent image based on the tracking results of the object in a past image, and the areas of each of the plurality of object candidates.

8. An object tracking device according to claim 6, wherein, when the tracking unit cannot identify the tracking target based on the tracking prediction, the tracking unit identifies the tracking target based on a comparison between the specified object information and information based on a plurality of the object candidates.

9. An object tracking method that extracts multiple object candidates for one object corresponding to specified object information from a time series of images, identifies a tracking target from the multiple object candidates based on object tracking results for past images, and tracks the tracking target.

10. An object tracking method according to claim 9, wherein a plurality of parameter values ​​are generated to be set when performing the process of extracting the object candidates, and a plurality of object candidates are extracted by respectively setting the values ​​of the plurality of parameters.

11. An object tracking method according to claim 10, wherein a plurality of values ​​of the parameter that sets the degree to which the object candidate is extracted is generated, and a plurality of the object candidates are extracted by respectively setting a plurality of the parameter values.

12. The object tracking method according to claim 9, wherein the tracking target is identified from a plurality of object candidates based on tracking prediction for a future image based on object tracking results for past images.

13. An object tracking method according to claim 12, wherein the tracking target is identified based on a comparison between the tracking prediction, which represents a predicted object area in a subsequent image based on the object tracking results for a past image, and the areas of each of a plurality of object candidates.

14. An object tracking method according to claim 12, wherein, when the tracking target cannot be identified based on the tracking prediction, the tracking target is identified based on a comparison between the specified object information and information based on a plurality of the object candidates.

15. A computer-readable storage medium storing a program that causes a computer to execute the following process: extracting multiple object candidates for one object corresponding to specified object information from a time series of images, identifying a tracking target from the multiple object candidates based on the object tracking results for past images, and tracking the tracking target.

Images