Target recognition system, target recognition method, and target recognition program product

Abstract

The disclosure relates to a target recognition system, a target recognition method, and a target recognition program product. The present disclosure provides a target object recognition technique that reduces the processing load when recognizing a target object presented in an image captured by an infrastructure camera. A target identification system identifies a target presented in an image captured by an infrastructure camera. A target recognition system detects a first target presented in an image as a temporary target. When the type of the temporary object is a moving object, the object recognition system applies a tracking process to the temporary object, and determines that the first object is an actually existing moving object on the basis of the detection result of the temporary object in the first period. In a case where the category of the temporary target is a stationary target, the target recognition system does not apply a tracking process to the temporary target, but determines that the first target is an actually existing stationary target based on a detection result of the temporary target in the second period.

Description

Technical Field

[0001] The present disclosure relates to object recognition technology for recognizing objects present in images captured by infrastructure cameras. Background Art

[0002] Patent Document 1 discloses a radar module that senses both stationary and moving objects. The radar module includes a stationary object processing unit that processes signals related to stationary objects, and a moving object processing unit that processes signals related to moving objects. The radar module also includes a switching unit that switches between processing by the stationary object processing unit and processing by the moving object processing unit. The radar module senses stationary and moving objects in different steps (i.e., at different timings).

[0003] Non-Patent Document 1 discloses a tracker called "ByteTrack".

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2021-143924

[0007] Non-patent literature

[0008] Non-Patent Literature 1: Zhang et al., “ByteTrack: Multi-Object Tracking by Associating Every Detection Box,” arXiv:2110.06864v3 [cs.CV], April 2022 (https: / / arxiv.org / abs / 2110.06864)

[0009] Consider identifying objects in images captured by infrastructure cameras. To prevent false object detection, one approach is to track objects detected within the image to confirm that they are actually present. However, tracking is complex, increasing the processing load and consuming computing resources. Summary of the Invention

[0010] One object of the present disclosure is to provide an object recognition technology that reduces the processing load when recognizing an object present in an image captured by an infrastructure camera.

[0011] The first aspect relates to an object recognition system for recognizing objects present in images captured by infrastructure cameras.

[0012] An object recognition system includes one or more processors. The one or more processors are configured to: detect a first object presented in an image as a temporary object; if the temporary object is a moving object, apply tracking processing to the temporary object and determine that the first object is an actual moving object based on detection results of the temporary object during a first period; and if the temporary object is a stationary object, not apply tracking processing to the temporary object and determine that the first object is an actual stationary object based on detection results of the temporary object during a second period.

[0013] The second viewpoint has the following viewpoints in addition to the first viewpoint.

[0014] The one or more processors output the result of determining that the first object is an actually existing moving object or an actually existing stationary object to a subsequent processing stage.

[0015] The third aspect relates to a method for recognizing an object that appears in an image captured by an infrastructure camera.

[0016] The object marker recognition method includes the following processing: detecting a first object marker presented in an image as a temporary object marker; if the temporary object marker is a moving object marker, applying tracking processing to the temporary object marker, and determining that the first object marker is an actually existing moving object marker based on a detection result of the temporary object marker in a first period; and if the temporary object marker is a stationary object marker, not applying tracking processing to the temporary object marker, but determining that the first object marker is an actually existing stationary object marker based on a detection result of the temporary object marker in a second period.

[0017] The fourth aspect relates to an object recognition program for recognizing objects present in images captured by infrastructure cameras.

[0018] The object recognition program is executed by a computer, causing the computer to perform the following processing: detecting a first object marker presented in an image as a temporary object marker; if the temporary object marker is a moving object marker, applying tracking processing to the temporary object marker, and determining that the first object marker is an actually existing moving object marker based on the detection result of the temporary object marker in a first period; and if the temporary object marker is a stationary object marker, not applying tracking processing to the temporary object marker, but determining that the first object marker is an actually existing stationary object marker based on the detection result of the temporary object marker in a second period.

[0019] Effects of the Invention

[0020] According to the first aspect, objects presented in an image are first detected as temporary objects. If the temporary object is classified as a moving object, tracking processing is performed to determine that the temporary object is actually a moving object. On the other hand, if the temporary object is classified as a stationary object, complex tracking processing is not required because the temporary object's position should remain unchanged within the image. Therefore, complex tracking processing is not performed and the temporary object is determined to be a stationary object. This reduces the processing burden and conserves computing resources compared to applying tracking processing to all objects regardless of their classification. This effect becomes more significant as the number of objects simultaneously presented in the image increases.

[0021] Furthermore, according to the second perspective, by outputting the recognition results of actual, confirmed object landmarks to subsequent processing, unnecessary processing and work associated with false detections can be suppressed. This helps reduce the processing burden on subsequent processing. Furthermore, frequent unnecessary processing and work (for example, false anomaly detections by street anomaly monitoring systems) can cause user anxiety and distrust. Therefore, the object landmark recognition system helps improve the reliability of subsequent processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a diagram explaining the outline of the tracking process.

[0023] Figure 2 This is a schematic diagram for explaining the tracking process at each time step.

[0024] Figure 3 This is a diagram for explaining the outline of the object mark recognition system according to this embodiment.

[0025] Figure 4 This is a schematic diagram for explaining the simplified determination process for each time step.

[0026] Figure 5 This is a schematic diagram for explaining an example where a determination condition actually exists.

[0027] Figure 6 This is a diagram illustrating another example of actual existence of a determination condition.

[0028] Figure 7 This is a block diagram showing a configuration example of an object recognition system.

[0029] Figure 8 This is a flowchart illustrating a series of processes related to the object recognition system.

[0030] Description of reference numerals:

[0031] 1: Object recognition system;

[0032] 10: Infrastructure cameras;

[0033] 11: Object detection department;

[0034] 12: tracker;

[0035] 110: processor;

[0036] 120: storage device;

[0037] 130: Communication device;

[0038] 200: Object recognition program;

[0039] 210: Object detection procedure;

[0040] 220: Tracking program;

[0041] BX: Bounding box. DETAILED DESCRIPTION

[0042] Embodiments of the present disclosure will be described with reference to the accompanying drawings.

[0043] 1. Object detection in images

[0044] A technology (object detection algorithm) for detecting objects in images captured by infrastructure cameras is known. This algorithm automatically detects objects in images by pre-learning the characteristics of various objects (objects to be detected). The object detection results are output to post-processing and utilized in various ways. Examples of post-processing include abnormality monitoring on streets and automated valet parking (AVP) control in parking lots.

[0045] However, the object detection algorithm may also cause the erroneous detection of objects that don't actually exist (hereinafter referred to as "false detections"). If a detection result containing a false detection is immediately output to a subsequent process (e.g., anomaly monitoring on a street), anomaly monitoring is performed based on the input detection result. If the detection result includes a falsely detected object (i.e., a false object that doesn't actually exist), anomalies may be detected for the false object. This is meaningless monitoring of an object that doesn't actually exist, resulting in a computational burden caused by unnecessary processing. Furthermore, repeated false detections can cause unnecessary anxiety for users of the anomaly monitoring system.

[0046] To prevent detection results containing false detections from being output to post-processing, it is necessary to confirm the actual presence of the detected object before outputting them to post-processing. For example, the object detection results of the object detection algorithm can be provisional. Object detection can then be continued for a certain period of time, and only when specific conditions are met can the detected object be confirmed to be actually present (confirmed presence). If the results after confirming presence are then output to post-processing, the likelihood of false detections being output to post-processing is reduced.

[0047] For example, the actual presence of an object can be confirmed through "tracking processing." Tracking processing refers to the process of continuously tracking an object detected in an image. The following describes tracking processing.

[0048] 1-1. Tracking and processing

[0049] Figure 1 This figure provides an overview of the tracking process. Images captured by the infrastructure camera 10 are sent to the object detection unit 11 as frames (still images constituting each frame). In this embodiment, the images formed by the connected frames constituting the image are defined as a series of images IMG. The object detection unit 11 uses an object detection algorithm to detect objects and their categories from the series of images IMG. A category refers to the type of object, with examples including vehicles, pedestrians, bicycles, traffic lights, and roadside trees. The object detection unit 11 further adds a bounding box BX around the detected objects. The bounding box BX indicates the position of the object within the series of images IMG. For example, the object detection unit 11 includes a machine learning model that has been trained to detect objects and their categories from the images IMG. By inputting the images IMG into the machine learning model, the object detection unit 11 can detect objects and their categories present in the images IMG and also add a bounding box BX around the objects. The object detector 11 sends information about the bounding boxes BX and object categories included in a series of images IMG to the tracker 12. The tracker 12 automatically tracks the same object within the series of images IMG using a tracking algorithm. For example, "ByteTrack" is known as a powerful tracker (see Non-Patent Document 1).

[0050] To track an object, the tracker 12 must continuously detect the object over a specified period. This is because if the object itself is not detected, it cannot be tracked. In other words, if tracking is performed properly on an object, the likelihood of its actual presence is high. On the other hand, if the object is not detected during the tracking process, the likelihood of its actual presence decreases. This allows the tracking process to confirm the actual presence of the object. If the result of confirming its actual presence is output as the final recognition result to subsequent processing, highly useful information is output in the subsequent processing.

[0051] Figure 2 This is a schematic diagram for explaining the tracking process at each time step. Figure 2 A series of images IMG at different time steps (t=t1, t2, t3, ...) included in the image are shown in FIG.

[0052] When an object moves, the bounding box BX representing it also moves within the series of images IMG. Multiple bounding boxes BX representing the same object within a series of images IMG at different time steps are spatially continuous. Therefore, by focusing on the movement of bounding boxes BX, multiple bounding boxes BX representing the same object within the series of images IMG can be identified. This allows tracking of the same object within the series of images IMG.

[0053] The tracker 12 (i.e., the tracking algorithm) tracks the same object in a series of images IMG based on the movement of the bounding box BX. More specifically, the tracker tracks the same object in a series of images IMG by determining multiple bounding boxes BXi representing the same object in the series of images IMG. Here, i (=1, 2, 3, ...) is an identifier of the multiple bounding boxes BX representing the same object. The tracker 12 associates the multiple bounding boxes BXi representing the same moving object in a series of images IMG at different time steps. It should be noted that the tracker 12 does not need to perform feature extraction in order to track the same object. The tracker 12 does not perform feature extraction, but tracks the same object based on the movement of the bounding box BX.

[0054] At t=t1, there are three object markers (O1 to O3) detected by the object marker detection unit 11. A corresponding bounding box BX1 to BX3 is added to each object marker. The object marker detection unit 11 also detects the type of each object marker. Figure 2In this case, the categories of objects O1 to O3 are pedestrians, bicycles, and roadside trees, respectively. Here, it is assumed that object O1 does not actually exist but is mistakenly detected. The tracker 12 predicts the movement of each bounding box BX. For example, the tracker 12 can use the position information of each bounding box BX before t = t1 to understand the movement trend of each bounding box BX. The tracker 12 can use this information to predict the detected position of each bounding box BX at the next time step (t = t2).

[0055] At t = t2 and t3, the tracker 12 can track bounding boxes BX2 and BX3. On the other hand, the tracker 12 cannot track bounding box BX1 at t = t2 and t3. Here, assume that the tracker 12 predicts that it can track bounding box BX1 at t = t2 based on the situation at t = t1. Despite this, bounding box BX1 is not detected. Therefore, the tracker 12 determines that object O1 corresponding to bounding box BX1 does not actually exist and that the object detection unit 11 has mistakenly detected object O1. In this way, the tracking process can accurately determine whether an object is misdetected.

[0056] However, tracking processing involves complex processes such as combining the positional information of each bounding box BX with a time series for prediction, and therefore tends to impose a high processing burden. In particular, when multiple objects appear within an image, the processing burden increases significantly when tracking these multiple objects simultaneously and in parallel. Therefore, this embodiment discloses object recognition technology that reduces the processing burden when recognizing objects in images captured by infrastructure cameras.

[0057] 2. Object recognition system disclosed herein

[0058] 2-1. Overview

[0059] The object recognition system 1 of the present disclosure categorizes detected objects as "moving objects" or "stationary objects" based on their categories. It then determines whether the objects are actual objects or erroneously detected by not applying tracking processing to stationary objects.

[0060] Figure 3 This is a diagram for explaining the outline of the object mark recognition system 1 of this embodiment. The object mark detection unit 11 performs Figure 1 The same process is repeated. Specifically, the object detector 11 detects a certain object (also referred to as the first object) and its category within the series of images IMG and assigns it a bounding box BX. False object detections may occur during this stage. For ease of explanation, the objects detected by the object detector 11 are referred to as "temporary objects."

[0061] Temporary objects detected by the object detection unit 11 in the object recognition system 1 are classified as either "moving objects" or "stationary objects." A moving object is one that can move, while a stationary object is one that cannot move. Examples of moving objects include vehicles, pedestrians, bicycles, and animals. On the other hand, examples of stationary objects include traffic lights, roadside trees, poles, and guardrails. It should be noted that whether an object is moving or stationary is determined not by whether it is moving or stationary at the time of detection by the object detection unit 11, but rather by whether the object possesses the properties of being movable. For example, even if a vehicle is parked when the object detection unit 11 detects it, it is still classified as a moving object because it is a movable object. The object detection unit 11 learns object categories (such as cars and pedestrians) by associating them with classes (moving objects / stationary objects). Therefore, object categories also include information about moving objects and stationary objects. For ease of explanation, among the temporary objects detected by the object detection unit 11, those that have been classified as moving objects are referred to as "temporary moving objects," and those that have been classified as stationary objects are referred to as "temporary stationary objects."

[0062] When viewed from the infrastructure camera 10, temporary stationary objects should not move. In other words, their positions should remain constant within the image IMG. Therefore, complex tracking processing is not necessarily required for temporary stationary objects. Based on this perspective, the object recognition system 1 determines whether a temporary object is actually present by performing different processing on temporary moving objects and temporary stationary objects. The object recognition system 1 applies the aforementioned tracking processing to temporary moving objects. The essential component of the tracking processing is the tracker 12. On the other hand, the object recognition system 1 performs the "simple determination processing" described below on temporary stationary objects instead of tracking. The essential component of the simple determination processing is the object detection unit 11. Information regarding temporary objects determined to be actually present through the tracking processing or the simple determination processing is output to subsequent processing.

[0063] 2-2. Simple judgment processing

[0064] The simplified determination process determines whether a temporary stationary object actually exists without applying tracking. Specifically, the object detection unit 11 performs further detection of temporary stationary objects for a predetermined period after detecting a temporary stationary object. Based on this detection result, the object detection unit 11 determines whether the temporary stationary object actually exists. The object detection unit 11 determines whether a temporary stationary object actually exists based on whether a temporary stationary object detected once is subsequently detected at substantially the same location. Since the simplified determination process does not utilize complex tracking algorithms, the processing burden can be reduced compared to tracking. The object recognition system 1 confirms that a temporary stationary object determined to be actually present by the simplified determination process is an actual object and outputs the recognition result confirming its actual presence to subsequent processing.

[0065] Figure 4 This is a schematic diagram for explaining the simplified determination process at each time step. Figure 2 Likewise, Figure 4 A series of images IMG at different time steps (t=t1, t2, t3, ...) included in the image are shown in FIG.

[0066] At t = t1, the object and category detection by the object detection unit 11 is complete. Each object is assigned a bounding box BX and a "moving" or "stationary" display. A pedestrian (moving object) is assigned a bounding box BX4. This object (pedestrian) is referred to as a temporary moving object Om4. A roadside tree (stationary object) is assigned a bounding box BX5. This object (roadside tree) is referred to as a temporary stationary object Os5. A traffic light (stationary object) is assigned a bounding box BX6. This object (traffic light) is referred to as a temporary stationary object Os6. It is assumed here that the temporary stationary object Os6 does not actually exist (i.e., it was mistakenly detected by the object detection unit 11).

[0067] and Figure 2 Similarly, the tracking process is applied to the temporary moving object Om4. Figure 2 Similarly, there is a possibility of misdetection of the object O1 in . Figure 4 In the process, the temporary moving object Om4 is continuously tracked until t = t3. Therefore, it can be determined that the temporary moving object Om4 actually exists.

[0068] Temporary stationary object marker Os5 was continuously detected at substantially the same position within image IMG until t=t3. Continuous detection of a temporary stationary object marker at substantially the same position over a predetermined period, as in the case of temporary stationary object marker Os5, is an example of a condition for determining actual presence. Therefore, the object detector 11 determines that the temporary stationary object marker is an actual presence marker. Details and other examples of the actual presence determination condition will be described later.

[0069] On the other hand, the temporary stationary object marker Os6 is not detected at t = t2 and t = t3. In this case, the actual presence determination condition is deemed not to be met. Therefore, the object detector 11 determines that the temporary stationary object marker Os6 was detected despite not actually existing (i.e., a false detection).

[0070] 2-3. Actual existence determination conditions

[0071] The actual existence determination conditions (conditions for determining that a temporary stationary object is an actually existing stationary object) will be described below.

[0072] An example of the actual presence determination condition is “a temporary still object marker is detected at substantially the same position in the image IMG for X consecutive frames (X is an integer greater than or equal to 2)”. Figure 5 This is a schematic diagram for explaining an example of the actual existence determination condition where X=5. Figure 5 1 is a diagram showing five frames extracted from a series of images IMG. In this diagram, a temporary stationary object Os5 and a pedestrian are depicted.

[0073] exist Figure 5 In the image, the pedestrian approaches the temporary stationary object Os5 but does not overlap it. In other words, the pedestrian is not hiding the temporary stationary object Os5 from the field of view of the infrastructure camera 10. Therefore, the temporary stationary object Os5 is detected at substantially the same position within the image IMG for five consecutive frames. Therefore, the object detection unit 11 determines that the temporary stationary object Os5 is a real object. In this way, when a temporary stationary object is detected for multiple consecutive frames, the object detection unit 11 determines that the temporary stationary object is a real object.

[0074] Another example of the actual presence determination condition is that “the temporary still object marker is detected Y (Y is an integer equal to or greater than 2) times at substantially the same position in the image IMG during a predetermined period”. Figure 6 This is a diagram illustrating another example of the actual existence of a certain condition (similar to Figure 5Similarly, 5 frames of a series of images IMG are extracted.) At the second frame, most of the temporary stationary object marker Os5 is hidden from the field of view of the infrastructure camera 10 by the passing vehicles. Therefore, the object marker detection unit 11 cannot continuously detect the temporary stationary object marker Os5 (not assigned a bounding box BX). In addition, at the fourth frame, the temporary stationary object marker Os5 is not hidden by other objects, but is also not detected. For example, due to the temporary light environment, etc., the object marker detection unit 11 is in a state where it cannot properly detect the temporary stationary object marker Os5 (detection omission). In the case of such a temporary detection omission, it is inappropriate to determine that the temporary stationary object marker Os5 does not actually exist (false detection). Therefore, in the case where a temporary stationary object marker is detected multiple times within a specified period, it is appropriate for the object marker detection unit 11 to determine that the temporary stationary object marker is an actually existing object marker. For example, in the case where the actual existence determination condition is "detected more than three times in 5 frames", in Figure 6 In the case of , the temporary stationary object mark Os5 is detected three times in five frames, and therefore the object mark detection unit 11 determines that the temporary stationary object mark Os5 is an actual object mark.

[0075] The actual presence determination conditions described above can also be used for temporary moving objects, that is, as actual presence determination conditions in the tracking process. In this case, simply replace "temporary stationary object" with "temporary moving object" in the above conditions. Furthermore, the "predetermined period" used in determining the actual presence determination conditions can be different for temporary stationary objects and temporary moving objects. Specifically, a "first period" can be set for temporary moving objects, while a "second period" can be set for temporary stationary objects.

[0076] 2-4. Effect

[0077] As described above, the object recognition system 1 performs a simplified determination process on temporary stationary objects instead of tracking them. Consequently, the object recognition system 1 can reduce the processing burden compared to applying tracking to all objects regardless of their type. In particular, applying tracking to multiple objects simultaneously and in parallel significantly increases the processing burden. Therefore, utilizing the object recognition system 1 is particularly effective when multiple objects are present within a series of images IMG.

[0078] Furthermore, by outputting the recognition results of objects that meet the actual presence determination criteria to subsequent processing, unnecessary processing and work in subsequent processing can be suppressed. This helps reduce the processing burden of subsequent processing. Furthermore, frequent unnecessary processing and work (for example, false anomaly detections by street anomaly monitoring systems) can cause user anxiety and distrust. Therefore, the object recognition system 1 outputs the results of the simplified determination process to subsequent processing, thereby helping to improve the reliability of subsequent processing.

[0079] 3. Example of composition

[0080] 3-1. Example of structure

[0081] Figure 7 1 is a block diagram showing a configuration example of the object recognition system 1. The object recognition system 1 includes one or more processors 110 (hereinafter simply referred to as “processors 110”), one or more storage devices 120 (hereinafter simply referred to as “storage devices 120”), and a communication device 130.

[0082] Processor 110 performs various processes. For example, processor 110 includes a CPU (Central Processing Unit). Storage device 120 stores various information required for processing. Examples of storage device 120 include volatile memory, nonvolatile memory, HDD (Hard Disk Drive), and SSD (Solid State Drive).

[0083] The object recognition program 200 is a computer program that performs object recognition processing. The object recognition program 200 is stored in the storage device 120. Alternatively, the object recognition program 200 may be recorded on a computer-readable recording medium. The object recognition program 200 is executed by the processor 110. The processor 110, executing the object recognition program 200, collaborates with the storage device 120 to implement the functions of the object recognition system 1.

[0084] The processor 110 communicates with the infrastructure camera 10 via the communication device 130 to acquire a series of images IMG. Furthermore, the processor 110 performs object detection, tracking, and simple determination processing. Specifically, the processor 110 functions as the object detector 11 and tracker 12, with the object detector 11 and tracker 12 being the primary processors of each process. The object detection program 210 used for object detection and the tracking program 220 used for tracking are stored in the storage device 120. The object detection program 210 and tracking program 220 may also be stored on a computer-readable recording medium.

[0085] 3-2. Processing Flow

[0086] Figure 8 This is a flowchart for explaining a series of processes related to the object recognition system 1 .

[0087] In step S10, the processor 110 acquires a series of images IMG from the infrastructure camera 10 via the communication device 130. Thereafter, the process proceeds to step S20.

[0088] In step S20 , the processor 110 detects objects and categories in a series of images IMG and assigns a bounding box BX to each object. The process then proceeds to step S30 .

[0089] In step S30, processor 110 refers to the type of the temporary object detected in step S20. If the temporary object is a stationary object (step S30: Yes), the process proceeds to step S41. On the other hand, if the temporary object is not a stationary object (i.e., a moving object) (step S30: No), the process proceeds to step S42.

[0090] In step S41, the processor 110 performs a simplified determination process on a temporary object (in this case, a temporary stationary object). The process then proceeds to step S51.

[0091] In step S42, the processor 110 applies tracking processing to the temporary object (in this case, the temporary moving object). Thereafter, the process proceeds to step S52.

[0092] In steps S51 and S52, processor 110 determines whether the actual presence determination condition is satisfied. If the actual presence determination condition is satisfied (step S51 or S52: Yes), processing proceeds to step S70. On the other hand, if the actual presence determination condition is not satisfied (step S51 or S52: No), processing proceeds to step S60.

[0093] In step S60, processor 110 determines that the temporary object detected in step S20 is an erroneously detected object. The process then ends. Information about a temporary object determined to be an erroneously detected object is not output to subsequent processing, thereby reducing unnecessary processing and effort in the subsequent processing.

[0094] In step S70, the processor 110 determines that the temporary object detected in step S20 is an actually existing object (actual existence determination). The process then proceeds to step S80.

[0095] In step S80, the processor 110 outputs the recognition result of the actual presence confirmation to the subsequent processing. Since the result does not include objects for which the actual presence confirmation condition is not met, unnecessary processing and system work in the subsequent processing can be suppressed.

Claims

1. An object recognition system for recognizing objects in images captured by infrastructure cameras, wherein: The object recognition system includes one or more processors. The one or more processors are configured to: detecting a first object presented in the image as a temporary object; If the temporary object is a moving object, tracking the temporary object is performed, and determining that the first object is an actually existing moving object based on a detection result of the temporary object in a first period. as well as When the category of the temporary object is a stationary object, the tracking process is not applied to the temporary object, but the first object is determined to be an actually existing stationary object based on the detection result of the temporary object in the second period.

2. The object recognition system according to claim 1, wherein: The second period is a period of X frames, where X is an integer greater than or equal to 2. When the category of the temporary object is the stationary object and the temporary object is detected in the X consecutive frames, the one or more processors determine that the first object is the actually existing stationary object.

3. The object recognition system according to claim 1, wherein: When the category of the temporary object is the stationary object and the temporary object is detected Y times or more during the second period, the one or more processors determine that the first object is the actually existing stationary object, where Y is an integer greater than or equal to 2.

4. The object mark recognition system according to any one of claims 1 to 3, wherein: The one or more processors output a result of determining that the first object is the actually existing moving object or the actually existing stationary object to a subsequent processing stage.

5. A method for identifying an object in an image captured by an infrastructure camera, wherein: The object recognition method includes the following processing: detecting a first object presented in the image as a temporary object; If the temporary object is a moving object, tracking the temporary object is performed, and determining that the first object is an actually existing moving object based on a detection result of the temporary object in a first period. as well as When the category of the temporary object is a stationary object, the tracking process is not applied to the temporary object, but the first object is determined to be an actually existing stationary object based on the detection result of the temporary object in the second period.

6. An object recognition program product, comprising an object recognition program, wherein the object recognition program recognizes objects presented in an image captured by an infrastructure camera, The object recognition program is executed by a computer, causing the computer to perform the following processing: detecting a first object presented in the image as a temporary object; If the temporary object is a moving object, tracking the temporary object is performed, and determining that the first object is an actually existing moving object based on a detection result of the temporary object in a first period. as well as When the category of the temporary object is a stationary object, the tracking process is not applied to the temporary object, but the first object is determined to be an actually existing stationary object based on the detection result of the temporary object in the second period.

Images