Information processing program, information processing method, information processing device, detection program, detection method, and detection device

Abstract

The objective is to improve the detection accuracy of object detection. [Solution] The information processing device detects the bounding boxes of an object in multiple frames captured by a camera installed at a predetermined location. Based on the confidence level of each of the detected bounding boxes, the information processing device generates a heat map that represents the average confidence level at each coordinate within the area captured by the camera.

Description

Technical Field

[0006] , , , ,

[0005]

[0001] The present invention relates to an information processing program, an information processing method, an information processing apparatus, a detection program, a detection method, and a detection apparatus.

Background Art

[0002] In recent years, a technique for detecting objects and people existing in an image or video using a machine learning model has been known. For example, this technique detects a bounding box indicating the region of an object included in an image and a category (for example, a person).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, depending on conditions such as the position of a person and the distance from a camera, there are image positions where the machine learning model is good at detection and image positions where it is not good at detection, so the detection accuracy may decrease.

[0006] For example, even when performing person detection at the same time of day, the confidence level for detecting a person facing directly (0 degrees) at a distance of 5m from the camera may be 0.96, while the confidence level for detecting a person rotated 270 degrees at the same distance of 5m from the camera may be 0.59. Furthermore, even for a person rotated 315 degrees at a distance of 10m from the camera, the confidence level for detection may be lower in the morning than at night. Thus, there are blind spots, which are conditions under which person detection using machine learning models is difficult.

[0007] In one aspect, the objective is to provide an information processing program, information processing method, information processing device, detection program, detection method, and detection device that can improve the detection accuracy of object detection. [Means for solving the problem]

[0008] In the first proposal, the information processing program causes the computer to perform the following processes: detect the bounding boxes of an object in multiple frames captured by a camera installed at a predetermined location, and generate a heat map that represents the average confidence level at each coordinate within the range captured by the camera, based on the confidence level of each of the detected bounding boxes. [Effects of the Invention]

[0009] In one respect, it can improve the accuracy of object detection. [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 is a diagram illustrating the information processing device according to Example 1. [Figure 2] Figure 2 is a diagram illustrating the information processing device according to Example 1. [Figure 3] Figure 3 is a functional block diagram showing the functional configuration of the information processing device according to Embodiment 1. [Figure 4] Figure 4 is a diagram illustrating the detection process. [Figure 5]Figure 5 illustrates the generation of a heatmap. [Figure 6] Figure 6 is a diagram illustrating the correction of confidence levels. [Figure 7] Figure 7 is a flowchart showing the flow of the heat map generation process. [Figure 8] Figure 8 is a flowchart showing the flow of the person detection process. [Figure 9] Figure 9 illustrates an example of a hardware configuration. [Modes for carrying out the invention]

[0011] The following describes in detail, with reference to the drawings, embodiments of the information processing program, information processing method, information processing device, detection program, detection method, and detection device disclosed in this application. However, the present invention is not limited by these embodiments. Each embodiment can be combined as appropriate within a consistent scope. [Examples]

[0012] (Description of the information processing device 10) Figures 1 and 2 illustrate the information processing device 10 according to Embodiment 1. The information processing device 10 shown in Figure 1 is connected to a fixed-point surveillance camera 1 (hereinafter sometimes simply referred to as camera 1) installed at a predetermined location.

[0013] Camera 1 captures video data (sometimes simply referred to as "video" below) within a certain range and outputs it to the information processing device 10. The size and number of pixels of each frame included in the video data captured by Camera 1 are constant. The information processing device 10 is an example of a computer that uses a pre-trained detection model to detect people in each frame (image data) that makes up the video input from Camera 1, and detects suspicious persons.

[0014] Generally, in person detection using a machine learning model, there is a blind spot which is a condition where detection is difficult. Therefore, for example, if a malicious third party knows the blind spot, they may perform actions such as "move so as not to face the surveillance camera" or "move in the morning when detection is difficult" to avoid person tracking. As a result, there is a risk of false negatives in detecting suspicious persons.

[0015] Therefore, the information processing apparatus 10 according to Example 1 corrects the confidence level at the time of person detection according to the detection position on the frame, for example, by increasing the detection confidence level at positions where person detection is difficult, thereby suppressing detection misses and improving the detection accuracy of person detection.

[0016] Here, the information processing apparatus 10 executes a process of generating a heat map used for correcting the confidence level, which is an example of the reliability, for each pixel of the image captured by the camera 1, and a process of correcting the confidence level at the time of person detection using the generated heat map. Therefore, the description will be divided into a generation phase for generating the heat map and a correction phase for correcting the confidence level at the time of person detection.

[0017] (Generation Phase) First, the process of the information processing apparatus 10 generating a heat map will be described using FIG. 1. Specifically, the information processing apparatus 10 detects bounding boxes (hereinafter sometimes referred to as "BB" or "bb") of persons in a plurality of frames captured by the camera 1 installed at a predetermined point. Then, the information processing apparatus 10 generates a heat map representing the average confidence level at each coordinate in the range captured by the camera 1 based on the confidence level of each of the detected plurality of bounding boxes.

[0018] For example, as shown in Figure 1, the information processing device 10 inputs each captured image (each frame) captured by the camera 1 over a predetermined period (e.g., 24 hours, morning, noon, and night, one week, etc.) into a detection model to obtain detection results including bounding box (BB) detection and confidence level. The information processing device 10 then calculates the average confidence level for each pixel in each frame. Subsequently, the information processing device 10 generates a heatmap that associates each pixel in the imaging range with the average confidence level.

[0019] Next, using Figure 2, we will explain the process by which the information processing device 10 corrects the confidence level using a heatmap. Specifically, the information processing device 10 uses a detection model to detect the bounding box of a person and the confidence level of person detection for frames of the target to be detected captured by the camera 1 installed at a predetermined location. Subsequently, the information processing device 10 corrects the confidence level detected from the frames of the target to be detected using a heatmap that is generated based on multiple frames captured by the camera 1 and represents the average confidence level calculated based on the confidence levels of multiple bounding boxes detected based on the multiple frames.

[0020] For example, as shown in Figure 2, after generating a heatmap, the information processing device 10 inputs the captured image taken by the camera 1 into the detection model and obtains a detection result including the broad beam (BB) and confidence level. The information processing device 10 then obtains the average confidence level of the region (multiple pixels) corresponding to the BB in the detection result from each region of the heatmap, and corrects the confidence level of the detection result with the average confidence level. Subsequently, the information processing device 10 determines that a person has been detected if, for example, the corrected confidence level is above a threshold, and determines that no person has been detected if the corrected confidence level is below the threshold.

[0021] In this way, the information processing device 10 can improve the detection success rate (true positive) even in cases where a false negative is likely to occur, thus achieving a "safety first" approach.

[0022] (Functional Configuration) Figure 3 is a functional block diagram showing the functional configuration of the information processing device 10 according to Embodiment 1. As shown in Figure 3, the information processing device 10 includes a communication unit 11, an output unit 12, a storage unit 13, and a control unit 20.

[0023] The communication unit 11 is a processing unit that controls communication with other devices, and is implemented, for example, by a communication interface. For example, the communication unit 11 receives video data containing multiple frames from the camera 1.

[0024] The output unit 12 is a processing unit that displays various information, and is implemented by, for example, a display or a touch panel. For example, the output unit 12 may display the generated heat map, the detection result before correction, or the final detection result after correction.

[0025] The storage unit 13 is a processing unit that stores various data and programs executed by the control unit 20, and is implemented by, for example, memory or a hard disk. This storage unit 13 stores the detection model 14, the imaging data DB (Database) 15, and the heatmap DB 16.

[0026] Detection model 14 is a pre-trained machine learning model that outputs a person detection result in response to an image (frame) input. For example, detection model 14 outputs "BB (category), confidence level" as the person detection result. Here, BB includes the region determined to contain a person and the detected category (person), and the confidence level indicates the confidence that a person is included in the region indicated by BB. The confidence level is expressed as a value in the range of "0.0" to "1.0", with a value closer to "1.0" indicating higher confidence.

[0027] The imaging data DB15 is a database that stores video footage, including multiple frames, captured by camera 1. For example, the imaging data DB15 stores frames (images) that capture the imaging range of camera 1, along with the shooting conditions such as the time, date and time, time of day, season, time of year, and weather when the frame was captured.

[0028] The heatmap DB16 is a database that stores the heatmap H generated by the heatmap generation unit 30, which will be described later. For example, the heatmap DB16 stores a heatmap H that represents the average confidence level at each coordinate in the area captured by the camera 1, based on the confidence level of each of the multiple bounding boxes detected from the captured image.

[0029] The control unit 20 is the processing unit that oversees the entire information processing device 10, and is implemented by, for example, a processor. This control unit 20 has a heat map generation unit 30 and a detection processing unit 40. The heat map generation unit 30 and the detection processing unit 40 are implemented by electronic circuits of the processor or processes executed by the processor.

[0030] The heatmap generation unit 30 comprises a detection unit 31 and a generation unit 32, and is a processing unit that generates a heatmap H used to correct the confidence level of the person detection result and stores it in the heatmap DB 16. The heatmap generation unit 30 can generate heatmap H for any period, such as by season, time of year, or weather.

[0031] The detection unit 31 is a processing unit that performs person detection using the detection model 14 on images (frames) captured by the camera 1. Specifically, the detection unit 31 performs person detection on each time frame stored in the imaging data DB 15 and outputs the detection result to the generation unit 32.

[0032] Figure 4 is a diagram illustrating the detection process. As shown in Figure 4, the detection unit 31 inputs an image captured as an example of a frame into the detection model 14 and obtains the detection result from the detection model 14. Here, the detection result includes BB (category = person) which contains a person, and a confidence level α.

[0033] The generation unit 32 is a processing unit that generates a heatmap H using the person detection results of each captured image by the detection unit 31. Specifically, the generation unit 32 calculates the average confidence level of the bounding box of at least one person containing the coordinate for each coordinate in the imaging range (frame). Then, the generation unit 32 generates a heatmap H that associates the average confidence level with each coordinate in the imaging range. Note that each coordinate can be read as each pixel or each pixel coordinate, etc.

[0034] Figure 5 illustrates the generation of heatmap H. Figure 5 shows an example of generating a heatmap using detection results T0, T1, T2, T3, and T4, which are detection results for each captured image with different shooting conditions, such as time of day. Each detection result is assumed to consist of pixels from (x,y)=(0,0) to (x,y)=(9,7), where the width of the image is the x-axis and the height of the image is the y-axis.

[0035] As shown in Figure 5, first, the generation unit 32 obtains bb1 (person A1, confidence level α1) and bb2 (person B1, confidence level β1) from the detection result T0, and obtains bb3 (person A2, confidence level α2) from the detection result T1. Similarly, the generation unit 32 obtains bb4 (person B2, confidence level β2) from the detection result T2, obtains bb5 (person B3, confidence level β3) and bb6 (person A3, confidence level α3) from the detection result T3, and obtains bb7 (person B4, confidence level β4) from the detection result T4.

[0036] Next, the generation unit 32 calculates the average confidence score of bb containing the pixels from (x,y)=(0,0) to (x,y)=(9,7). For example, let's consider pixel (P) corresponding to (x,y)=(7,5). In this case, the generation unit 32 obtains the confidence score α1 for bb1 containing pixel (P), the confidence score α2 for bb3, and the confidence score α3 for bb6. Then, the generation unit 32 calculates "(confidence score α1 + confidence score α2 + confidence score α3) / 3" as the average confidence score of pixel (P).

[0037] Let's take the pixel (Q) corresponding to (x,y)=(2,3) as an example. In this case, the generation unit 32 obtains the confidence values ​​β1 for bb2, β2 for bb4, β3 for bb5, and β4 for bb7, which contain pixel (Q). Then, the generation unit 32 calculates "(confidence value β1 + confidence value β2 + confidence value β3 + confidence value β4) / 4" as the average confidence value of pixel (Q).

[0038] In this way, the generation unit 32 calculates the average confidence level for each pixel of the image captured by the camera 1 and generates a heatmap H that associates the confidence level with each pixel. For example, in the heatmap H of the video from the fixed-point surveillance camera 1 shown in Figure 5, the average confidence level of each pixel in region G is high, indicating that the detection accuracy of this region is high, while the average confidence level is low near the center, indicating that the detection accuracy is low.

[0039] Returning to Figure 3, the detection processing unit 40 comprises a detection unit 41, a correction unit 42, and a determination unit 43, and is a processing unit that performs person detection from the captured image of the target to be detected using a heat map H.

[0040] The detection unit 41 is a processing unit that performs person detection using the detection model 14 on the image (frame) to be detected, which is captured by the camera 1. The processing of the detection unit 41 is the same as in Figure 4, so a detailed explanation is omitted. Here, as an example, it is assumed that the detection unit 41 obtains a detection result T including bb (person, confidence level 0.6) from the image to be detected and outputs it to the correction unit 42.

[0041] The correction unit 42 is a processing unit that corrects the confidence level of the detection result detected by the detection unit 41 using a heat map H. For example, the correction unit 42 calculates the average value of the average confidence level of each coordinate in the captured image (imaging range) that is included in the bounding box of the person detected from the frame to be detected. Then, the correction unit 42 corrects the confidence level of the bounding box of the person detected from the frame to be detected by the average value of the average confidence level.

[0042] Figure 6 illustrates the correction of the confidence level. As shown in Figure 6, the correction unit 42 obtains a detection result T that includes bb (person, confidence level 0.6). Here, bb is a region that includes a total of 6 pixels: (2,5), (2,6), (2,7), (3,5), (3,6), (3,7).

[0043] Next, the correction unit 42 identifies the region of heatmap H corresponding to bb "(2,5), (2,6), (2,7), (3,5), (3,6), (3,7)". Then, the correction unit 42 obtains the average confidence scores "C1, C2, C3, C4, C5, C6" corresponding to each pixel in the region of heatmap H "(2,5), (2,6), (2,7), (3,5), (3,6), (3,7)". Subsequently, the correction unit 42 calculates the average of the obtained average confidence scores for each pixel using "(C1+C2+C3+C4+C5+C6) / 6". Here, it is assumed that the average of the average confidence scores is calculated to be "0.7".

[0044] Subsequently, the correction unit 42 corrects the confidence level of bb in the detection result T, which is "0.6", by the average value of the average confidence levels, which is "0.7". For example, the correction unit 42 calculates the corrected confidence level of "0.6 / 0.7 ≈ 0.86" by dividing the detection result by the average value of the average confidence levels. Then, the correction unit 42 outputs the detection result T, which includes bb (person, corrected confidence level 0.86), to the judgment unit 43.

[0045] The correction unit 42 may also perform the correction process if the confidence level of bb in the detection result T is less than a threshold. Furthermore, the calculation method used by the correction unit 42 for correction is not limited to the method described in Figure 6, and pre-prepared functions may also be used.

[0046] The determination unit 43 is a processing unit that determines whether a person has been detected using the corrected confidence level, which has been corrected by the correction unit 42. For example, the determination unit 43 determines that a person is present if the corrected confidence level is equal to or greater than a threshold. In the example shown in Figure 6, the correction unit 42 determines that a person is present at bb in the detection result T because the corrected confidence level of "0.86" is equal to or greater than the threshold (0.8). In other words, the determination unit 43 determines that a person, such as a suspicious person, has been detected in the image to be detected.

[0047] (Heatmap generation process flow) Figure 7 is a flowchart showing the flow of the heatmap generation process. As shown in Figure 7, when the heatmap generation unit 30 is instructed to start processing by an administrator or the like (S101: Yes), it uses the detection model 14 to obtain, for example, the results of detecting people for each image taken in the past (S102).

[0048] Next, the heatmap generation unit 30 acquires the confidence level for each pixel in the imaging range (S103). Then, the heatmap generation unit 30 calculates the average confidence level (average confidence level) for each pixel containing the pixel (S104). After that, the heatmap generation unit 30 generates a heatmap H that associates the average confidence level with each pixel (S105).

[0049] (Flow of person detection process) Figure 8 is a flowchart showing the flow of the person detection process. As shown in Figure 8, when the detection processing unit 40 receives an image to be detected (S201: Yes), it inputs the image to the detection model 14 and performs person detection (S202).

[0050] Here, if no person (bb: bounding box) is detected (S203: No), the detection processing unit 40 determines that no person was detected in the image to be detected (S204).

[0051] On the other hand, if a person (bb) is detected (S203:Yes), the detection processing unit 40 determines whether the confidence level of the detected bb is above a threshold (S205).

[0052] Then, if the confidence level of the detected bb is above a threshold (S205: Yes), the detection processing unit 40 determines that a person has been detected in the image to be detected and outputs the person detection result, including information about bb (S206).

[0053] On the other hand, if the confidence level of the detected bb is less than a threshold (S205: No), the detection processing unit 40 obtains the average confidence level of each pixel corresponding to the area of ​​person detection (bb) from the heatmap (S207).

[0054] Next, the detection processing unit 40 calculates the average value of the average confidence score for each pixel (S208), and corrects the confidence score of the detected person (bb) using the average value of the average confidence scores (S209).

[0055] Then, if the confidence level of the corrected bb is above the threshold (S210: Yes), the detection processing unit 40 determines that a person has been detected in the image to be detected and outputs the person detection result, including the bb information (S206). On the other hand, if the confidence level of the corrected bb is below the threshold (S210: No), the detection processing unit 40 determines that no person has been detected in the image to be detected (S211).

[0056] (effect) As described above, the information processing device 10 generates a heatmap that represents the average confidence level at each coordinate within the range captured by the camera 1, based on the confidence levels of each of the multiple bbs. As a result, the information processing device 10 can provide visually accessible information showing a list of confidence levels, allowing for the determination of the reliability of the confidence levels during person detection and improving the detection accuracy of person detection.

[0057] Furthermore, by creating a heatmap, the information processing device 10 allows administrators to identify locations with low detection accuracy or locations where detection errors are likely to occur. The information processing device 10 can also provide administrators with information that enables them to find optimal camera placements with fewer weaknesses and suitable locations for camera 1.

[0058] Furthermore, when using an existing detection model in an operating environment different from that used during training, weaknesses (conditions resulting in low detection performance) due to environmental differences often occur. In such cases, fine-tuning the detection model to correct these weaknesses is sometimes necessary, but this is costly. However, by using the information processing device 10 according to Example 1, the weaknesses of the detection model can be corrected at a low cost.

[0059] Furthermore, the information processing device 10 performs person detection using a trained detection model and generates a heatmap using the person detection results, so it can generate a heatmap using an accurate confidence score.

[0060] Furthermore, the information processing device 10 can correct the confidence level of the person detection result using a heatmap and then make a final determination of person detection. As a result, the information processing device 10 can achieve accurate person detection even under conditions that the detection model struggles with, thereby reducing missed detections and improving the accuracy of person detection. In addition, the above processing can be used to propose a method for positioning cameras in the scene to make person detection more robust. [Examples]

[0061] Now, although embodiments of the present invention have been described, the present invention may be implemented in various other forms besides those described above.

[0062] (Numerical values, etc.) The image size, pixel count, numerical values, thresholds, etc., used in the above embodiment are merely examples and can be changed as needed. Furthermore, the processing flow described in each flowchart can be modified as appropriate within a consistent range. The information processing device 10 is an example of a detection device, and the detection model is an example of a machine learning model. While the above embodiment was explained using person detection as an example, it is not limited to this and can be similarly applied to object detection such as cars.

[0063] Furthermore, while the above embodiment describes an example of generating a heatmap using the confidence score output by the detection model, it is not limited to this. For example, the confidence score of each pixel in the image may be calculated based on past detection results, or the confidence score of each pixel in the image may be set by the user. Note that the same threshold can be used for comparing the confidence scores before and after correction.

[0064] (system) Unless otherwise specified, the processing procedures, control procedures, specific names, and various data and parameters shown in the above documents and drawings may be changed at will.

[0065] Furthermore, the specific forms of distribution and integration of the components of each device are not limited to those shown in the diagram. For example, the heatmap generation unit 30 and the detection processing unit 40 may be executed by separate devices. In other words, all or part of the components may be functionally or physically distributed and integrated in any unit depending on various loads and usage conditions. Moreover, all or any part of the processing functions of each device may be implemented by a CPU and a program executed for analysis on that CPU, or as hardware using wired logic.

[0066] Furthermore, each processing function performed by each device may be implemented, in whole or in part, by a CPU and a program executed for analysis by that CPU, or by wired logic hardware.

[0067] (Hardware) Figure 9 illustrates an example of a hardware configuration. As shown in Figure 9, the information processing device 10 includes a communication device 10a, an HDD (Hard Disk Drive) 10b, memory 10c, and a processor 10d. Furthermore, the components shown in Figure 9 are interconnected by a bus or the like.

[0068] The communication device 10a is a network interface card or the like, and communicates with other devices. The HDD 10b stores programs and databases that operate the functions shown in Figure 3.

[0069] The processor 10d operates a process that performs the functions described in Figure 3 by reading a program that performs the same processing as each processing unit shown in Figure 3 from the HDD 10b or the like and loading it into memory 10c. For example, this process performs the same functions as each processing unit of the information processing device 10. Specifically, the processor 10d reads a program that has the same functions as the heatmap generation unit 30 and the detection processing unit 40 from the HDD 10b or the like. Then, the processor 10d executes a process that performs the same processing as the heatmap generation unit 30 and the detection processing unit 40.

[0070] Thus, the information processing device 10 operates as an information processing device that executes information processing methods and detection methods by reading and executing a program. Furthermore, the information processing device 10 can also achieve the same functionality as the above-described embodiment by reading the program from the recording medium using a media reader and executing the read program. It should be noted that the program referred to in this other embodiment is not limited to being executed by the information processing device 10. For example, the above embodiment may also be applied to cases where another computer or server executes the program, or where these computers or servers collaborate to execute the program.

[0071] This program may be distributed via a network such as the Internet. Alternatively, this program may be recorded on a computer-readable recording medium such as a hard disk, flexible disk (FD), CD-ROM, MO (Magneto-Optical disk), or DVD (Digital Versatile Disc), and executed by being read from the recording medium by a computer. [Explanation of symbols]

[0072] 10 Information Processing Devices 11 Communications Department 12 Output section 13 Storage section 14 Detection Models 15. Imaging Data Database 16 Heatmap DB 20 Control Unit 30 Heatmap generation section 31 Detection unit 32 Generation part 40 Detection Processing Unit 41 Detection unit 42 Correction section 43 Judgment section

Claims

1. On the computer, The bounding box of an object is detected in multiple frames captured by a camera installed at a predetermined location. Based on the confidence level of each of the multiple bounding boxes detected, a heatmap is generated that represents the average confidence level at each coordinate within the area captured by the camera. An information processing program that executes a process.

2. The aforementioned detection process is, The process includes inputting each of the multiple frames, each of which is imaged within the aforementioned range at predetermined time intervals, into a machine learning model to detect the bounding box of the object and the confidence level, The information processing program according to claim 1.

3. The aforementioned generation process is, For each coordinate within the range, calculate the average confidence level of the bounding box of at least one object that includes the coordinate. A heatmap is generated in which the average confidence level is associated with each coordinate within the range. An information processing program according to claim 1 or 2, including processing.

4. Computers The bounding box of an object is detected in multiple frames captured by a camera installed at a predetermined location. Based on the confidence level of each of the multiple bounding boxes detected, a heatmap is generated that represents the average confidence level at each coordinate within the area captured by the camera. An information processing method that performs a process.

5. The bounding box of an object is detected in multiple frames captured by a camera installed at a predetermined location. Based on the confidence level of each of the multiple bounding boxes detected, a heatmap is generated that represents the average confidence level at each coordinate within the area captured by the camera. An information processing device having a control unit.

6. On the computer, A camera installed at a predetermined location captures frames of the object to be detected, and a predetermined machine learning model is used to detect the bounding box of the object and the confidence level of the object's detection. A heatmap generated based on multiple frames captured by the camera, the heatmap representing the average confidence level calculated based on the confidence levels of multiple bounding boxes detected based on the multiple frames, is used to correct the confidence level detected from the frame to be detected. A detection program that initiates a process.

7. The aforementioned heatmap is This information associates the average confidence level with each coordinate within the range captured by the camera. The aforementioned correction process is: The average value of the average confidence score for each coordinate within the imaging range that is included in the bounding box of the object detected from the frame of the object to be detected is calculated. The confidence level of the bounding box of the object detected from the frame to be detected is corrected by the average value. The detection program according to claim 6, including processing.

8. The aforementioned correction process is: If the confidence level of the bounding box of the object detected from the frame to be detected is less than a threshold, the confidence level is corrected to a value obtained by dividing it by the average value. The detection program according to claim 7, including processing.

9. If the corrected confidence level is above a threshold, it is detected that the object is within the bounding box of the object detected from the frame to be detected. The detection program according to claim 6 or 7, which causes the computer to perform the processing.

10. Computers A camera installed at a predetermined location captures frames of the object to be detected, and a predetermined machine learning model is used to detect the bounding box of the object and the confidence level of the object's detection. A heatmap generated based on multiple frames captured by the camera, the heatmap representing the average confidence level calculated based on the confidence levels of multiple bounding boxes detected based on the multiple frames, is used to correct the confidence level detected from the frame to be detected. A detection method for executing a process.

11. A camera installed at a predetermined location captures frames of the object to be detected, and a predetermined machine learning model is used to detect the bounding box of the object and the confidence level of the object's detection. A heatmap generated based on multiple frames captured by the camera, the heatmap representing the average confidence level calculated based on the confidence levels of multiple bounding boxes detected based on the multiple frames, is used to correct the confidence level detected from the frame to be detected. A detection device having a control unit.

Images