Attribute analysis system, and estimation method and communication method using the same
The surveillance camera system with a battery, solar panel, and AI server ensures power and stable communication, allowing accurate real-time analysis by correcting snapshot data, overcoming installation and data transmission challenges.
Patent Information
- Application Number
- JP2024039129
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2024-03-13
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-03-13
AI Technical Summary
Surveillance cameras require power and stable communication lines for operation, limiting their installation locations and facing challenges in real-time data transmission and analysis due to large video data sizes, which can lead to communication errors and reduced accuracy in AI analysis.
A surveillance camera system equipped with a battery and solar panel for power, a recording medium, and a communication unit with a SIM card, along with an AI server that includes video and image storage units and a central processing unit for analyzing images and videos, which performs statistical corrections to ensure accurate analysis even with communication errors.
Enables installation in power-constrained locations, stabilizes data transmission, and achieves accurate real-time analysis by correcting snapshot data to match video analysis accuracy, reducing analysis time and errors.
Smart Images

Figure 0007584694000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an attribute analysis system and an attribute analysis method using a surveillance camera, and in particular to an attribute analysis camera system and attribute analysis method thereof that can be installed regardless of the equipment environment such as a power source or optical fiber line and has reduced power consumption and communication volume. [Background technology]
[0002] The demand for systems that use surveillance cameras and AI (artificial intelligence) to analyze the number and attributes of subjects is increasing year by year. Such attribute analysis systems are used not only for crime prevention and deterrence and disaster prevention monitoring, but also for marketing and traffic volume surveys. Therefore, attribute analysis systems are required to be installed in various locations depending on their purpose.
[0003] As an example of an attribute analysis system, for example, Patent Document 1 and Patent Document 2 disclose a human attribute estimation system equipped with a surveillance camera. Here, Patent Document 1 discloses that in order to improve the accuracy of estimating the attributes of a person who is the subject of attribute estimation, regardless of the installation environment of the surveillance camera, estimation parameters based on learning data having environment-dependent attributes close to the environment-dependent attributes of the surveillance camera are used. Moreover, Patent Document 2 solves the problem of a decrease in the accuracy of attribute estimation caused by the illuminance of a person photographed by a surveillance camera being different from that of a sample image. Patent Document 2 describes how a pseudo-on-site image is generated as if a person had actually been photographed at the site, based on shooting environment data indicating the shooting environment under which the person to be the subject of attribute estimation is photographed, and a standard image which is a person image, and how an estimation model is trained using this pseudo-on-site image.
[0004] However, both Patent Document 1 and Patent Document 2 are premised on the use of surveillance cameras in stores. For this reason, neither Patent Document 1 nor Patent Document 2 describes stable communication of data from a surveillance camera when the surveillance camera is installed outdoors or in an open space where it is difficult to secure a power source, or a case where it is desired to analyze and output the degree of congestion in a certain time period (such as one hour) in real time, or a case where an analysis system is performed only using images specialized for more power-saving and low-volume data, a case where the number of people detected in a video in a certain time period (such as one hour) at the same time is output with a small error based on the number of people detected in a snapshot image (still image) taken by the surveillance camera. [Prior art documents] [Patent documents]
[0005] [Patent Document 1] JP 2012-252507 A [Patent Document 2] JP 2013-242825 A Summary of the Invention [Problem to be solved by the invention]
[0006] Depending on the purpose, surveillance cameras installed in locations to be monitored may need to be installed in locations where the environment, such as power supply and communication lines, is not well developed. However, surveillance cameras usually require a power supply to capture and transmit images. Therefore, the locations where surveillance cameras can be installed are limited to locations where power can be supplied. Also, for example, in the absence of an optical fiber line or other communication line, the attribute analysis system needs to communicate with the surveillance camera using a wireless line (LTE line). However, when communicating using such an LTE line, it may be difficult to stably transmit data from the surveillance camera to the analysis server because the data size of the video captured by the surveillance camera is large. In such cases, it may not be possible to perform analysis using highly accurate artificial intelligence (AI). Furthermore, when transmitting video, the large data size can lead to communication errors, which means that the AI (artificial intelligence) server cannot receive the video data necessary for analysis, and as a result, the AI server cannot perform attribute analysis. In addition, because analyzing a video requires a certain amount of time due to the data size, it is difficult to perform attribute analysis in a short period of time. [Means for solving the problem]
[0007] The present invention provides an attribute analysis system that solves the above problems. Specifically, the present invention relates to A surveillance camera including a power supply unit including a battery and a solar panel for charging the battery, a recording medium for storing a captured video and an image captured in a first period, and a communication unit for transmitting the video or the image recorded in the recording medium; An artificial intelligence server including a video storage unit that records the video received from the recording medium, an image storage unit that records the image received from the recording medium, and a central processing unit that analyzes the video or the image to detect people and analyzes the number of people and their attributes (gender, age), It is equipped with The artificial intelligence server, in the central processing unit, For the video and the images captured in advance multiple times at the same time during a predetermined period, the number of people detected from the video recorded in the video storage unit and the number of people detected from the images recorded in the image storage unit are calculated, and an approximation calculation is performed between the number of people detected from the video and the number of people detected from the images using a statistical method to calculate a correction value between the video and the images; The correction value is applied to the number of people detected in each of the images captured in the first period, the number of people in the images during a certain time period at the time the images were captured is calculated, and a real-time person count during the certain time period is simulated and analyzed. The present invention provides an attribute analysis system.
[0008] The present invention also provides a method for producing a semiconductor device comprising the steps of: A surveillance camera including a power supply unit including a battery and a solar panel for charging the battery, a recording medium for storing a captured video, and a communication unit for transmitting the video recorded on the recording medium; an artificial intelligence server including a video storage unit that records the video received from the recording medium, and a central processing unit that analyzes the video recorded in the video storage unit to detect people and determine the number of people; An attribute analysis system comprising: The artificial intelligence server includes: (a) causing the surveillance camera to transmit video images captured over a predetermined period of time to the video storage unit, and monitoring for occurrence of a communication error between the surveillance camera and the artificial intelligence server; (b) when the video is transmitted to the video storage unit without detecting the communication error, the central processing unit is caused to analyze the video, detect people, and calculate the number of people; (c) when the communication error is detected, causing the surveillance camera to transmit the video to the video storage unit again, and monitoring for the occurrence of the communication error between the surveillance camera and the artificial intelligence server during a first period; (d) repeating the above steps (a) to (c) for a second period longer than the first period. An attribute analysis system is also provided.
[0009] Here, it is preferable that the communication unit is equipped with a SIM card, or that the communication unit further includes an operation console that communicates with the artificial intelligence server via a wired or wireless connection using an Internet line.
[0010] Furthermore, the present invention also provides an estimation method in an attribute analysis system. in particular, A surveillance camera including a power supply unit including a battery and a solar panel for charging the battery, a recording medium for storing a captured video and an image captured in a first period, and a communication unit for transmitting the video or the image recorded in the recording medium; an artificial intelligence server including a video storage unit that records the video received from the recording medium, an image storage unit that records the images received from the recording medium, and a central processing unit that analyzes the video or the images to detect people and determine the number of people; An estimation method in an attribute analysis system comprising: The artificial intelligence server, in the central processing unit, a step of determining the number of people detected from the video recorded in the video storage unit and the number of people detected from the images recorded in the image storage unit for the video and the images taken multiple times at the same time during a predetermined period in advance, and performing an approximation calculation between the number of people detected from the video and the number of people detected from the images using a statistical method to determine a correction value between the video and the images; applying the correction value to the number of people detected in each of the images captured in the first period, to obtain the number of people at the time the image was captured; The present invention provides an estimation method in an attribute analysis system, comprising:
[0011] In addition, the present invention also provides a communication method in an attribute analysis system. in particular, A surveillance camera including a power supply unit having a battery and a solar panel for charging the battery, a recording medium for storing a captured video, and a communication unit for transmitting the video recorded on the recording medium; an artificial intelligence server including a video storage unit that records the video received from the recording medium, and a central processing unit that analyzes the video recorded in the video storage unit to detect people and analyzes the number of people and their attributes (gender, age), A communication method in an attribute analysis system comprising: The artificial intelligence server includes: (a') causing the surveillance camera to transmit video images captured over a predetermined period of time to the video storage unit, and monitoring for occurrence of a communication error between the surveillance camera and the artificial intelligence server; (b') when the video is transmitted to the video storage unit without detecting the communication error, causing the central processing unit to analyze the video, detect people, and calculate the number of people; (c') when the communication error is detected, causing the surveillance camera to transmit the video to the video storage unit again, and monitoring for the occurrence of the communication error between the surveillance camera and the artificial intelligence server during a first period; (d') repeating the steps (a') to (c') during a second period longer than the first period. The present invention provides a communication method in an attribute analysis system, which executes the above. Effect of the Invention
[0012] The surveillance camera according to the present invention is equipped with a charging module such as a solar panel and a battery. Therefore, the surveillance camera can be installed in a place where there is no power supply facility. The surveillance camera also has a communication unit with a built-in SIM (Subscriber Identity Module) card, for example. In other words, since there is no need for other devices to be installed together with the surveillance camera, the surveillance camera can be installed in a more space-saving place. The artificial intelligence server of the present invention employs a video acquisition algorithm that monitors communication errors between the surveillance camera and the artificial intelligence server during video transmission during video analysis, and requests the surveillance camera to resend data if a communication error occurs. This improves the reliability of video acquisition between the surveillance camera and the artificial intelligence server, allowing the artificial intelligence server to perform stable analysis. In addition, the artificial intelligence server according to the present invention can perform not only video analysis but also snapshot analysis, which analyzes images (whose data size is smaller than that of videos). Therefore, snapshot analysis can reduce the time required for analysis compared to video analysis, making it possible to realize analysis with less delay. Furthermore, the AI server according to the present invention performs statistical approximation calculations on the number of people found from the video and images obtained at the same time over the same period to find a correction value between the two, and applies the correction value to the number of people found by snapshot analysis. Therefore, the AI server according to the present invention can obtain accuracy in the number of people from snapshot analysis close to that of video analysis. Approximation calculation methods using statistical processing include methods using machine learning, methods using statistical methods including mathematical methods, etc. Examples of such statistical methods including mathematical methods include linear regression and state space models. [Brief description of the drawings]
[0013] [Figure 1] 1A is a block diagram showing the configuration of an attribute analysis system 1 of the present invention. 1B is a photograph showing a person detected by an artificial intelligence server 20 being marked. [Diagram 2] FIG. 13 is a data flow diagram showing the processing performed in the attribute analysis system 1 when a communication error occurs during communication of a moving image file executed during moving image analysis. [Diagram 3] 1 is a data flow diagram showing processing during snapshot analysis in the attribute analysis system 1. FIG. [Figure 4] 1 is a flow diagram showing a process of analyzing a moving image in the attribute analysis system 1. FIG. [Diagram 5] 1 is a flow diagram showing processing during snapshot analysis in the attribute analysis system 1. FIG. [Figure 6] The video files and image files acquired by the attribute analysis system 1 were analyzed over a two-month period, and the data was collected by tallying up the number of people who analyzed the video and the number of people who analyzed the snapshots. [Figure 7] 7 is a graph for finding an optimal correction coefficient between the number of people analyzed by video and the number of people analyzed by snapshot by performing regression analysis on the measurement data in FIG. 6 as an example of approximate calculation of statistical processing. [Figure 8] 8 is a graph of basic statistics obtained from the regression analysis shown in FIG. 7. [Figure 9] 8 is a graph showing the error (before correction) between the number of people who analyzed videos and the number of people who analyzed snapshots, and the error (after correction) when the correction value obtained from the regression analysis in FIG. 7 is used. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0014] With reference to FIG. 1A, an attribute analysis system 1 according to an embodiment of the present invention will be described. The attribute analysis system 1 includes a surveillance camera 10, an artificial intelligence server 20 communicating with the surveillance camera 10, and an operation console PC 30 communicating with the artificial intelligence server 20. Here, the surveillance camera 10 and the artificial intelligence server 20, and the artificial intelligence server 20 and the operation console PC 30 may be connected, for example, by a communication cable, wirelessly, or by both a wired and wireless connection. In particular, when the surveillance camera 10 is installed outdoors, it is preferable that the surveillance camera 10 and the artificial intelligence server 20 are connected at least wirelessly through an Internet line.
[0015] The surveillance camera 10 includes a power supply unit 101, a recording medium 102 that records videos (images) captured by the surveillance camera or images (still images) captured at a predetermined cycle (for example, a one-minute cycle as a first cycle), and a communication unit 103 that communicates with the artificial intelligence server 20. The power supply unit 101 includes a battery 101a and a solar panel 101b that charges the battery 101a. By including such a solar panel 101b, the surveillance camera 10 can be installed without being restricted by the power supply, for example, even outdoors where there is no power supply facility. The recording medium 102 may be any one of a hard disk drive, an SSD, and a memory card that has a capacity sufficient to record the captured videos or images. The communication unit 103 may include, for example, a built-in SIM card.
[0016] The artificial intelligence server 20 includes a recording unit 201 that records videos or images transmitted from the surveillance camera 10, and a central processing unit 202 that communicates with the recording unit 201 and performs calculations such as processing to analyze the videos or images. The recording unit 201 includes a video storage unit 201a that records videos and analysis programs from the surveillance camera 10, an image storage unit 201b that records images and analysis programs from the surveillance camera 10, and an analysis data recording unit 201c that records the results of analyzing the videos or images from the surveillance camera 10. The central processing unit 202 includes a moving object detection unit 202a, an artificial intelligence determination unit 202b, a marking unit 202c, a video analysis unit 202d, and a snapshot analysis unit 202e. Here, the moving object detection unit 202a, the artificial intelligence determination unit 202b, and the marking unit 202c are programs that analyze the videos captured by the surveillance camera 10 in cooperation with the central processing unit 202. The video analysis unit 202d is a program that performs video analysis in cooperation with the central processing unit 202. The snapshot analysis unit 202e is a program that performs video analysis in cooperation with the central processing unit 202. Here, the moving object detection unit 202a can detect a moving object by using, for example, image processing software called OpenCV, which is publicly available, and a frame difference method. In addition, the artificial intelligence determination unit 202b can determine whether the detected moving object is a person based on a model structure of a machine learning algorithm (for example, a Convolutional Neural Network model structure), and can determine the attributes (gender, age) of the person. When the artificial intelligence determination unit 202b determines that the moving object is a person, the marking unit 202c can execute a marking process that surrounds the person with a square frame, for example, as shown in FIG. 1B described later (note that the video analysis unit 202d and the snapshot analysis unit 202e will be described later). The results of the analysis by the artificial intelligence server 20 can be displayed on the operation console PC 30 connected to the artificial intelligence server 20 (for example, the operation console PC 30 can display a screen in which people detected from a video or image are individually marked, as shown in FIG. 1B described later). It is also possible to display the attributes of the people (gender, age).
[0017] Fig. 1B shows a screen when the central processing unit 202 analyzes a video or image and performs a marking process on a detected person. Such a screen can be displayed on the operation console PC 30. Here, Fig. 1B shows a case where the detected person is marked by surrounding each person with a square frame. However, it is also possible to mark the detected person in a different way (such as surrounding it with a circle). It is also possible to display the attributes (gender, age) of the person.
[0018] FIG. 2 shows a process of communicating several videos (video files) between the surveillance camera 10 and the artificial intelligence server 20 in video analysis. First, in S1, the artificial intelligence server 20 requests the surveillance camera 10 to transfer the video file to the video storage unit 201a, and monitors whether a communication error occurs between the surveillance camera 10 and the artificial intelligence server 20. Here, in S2, a communication error occurs between the surveillance camera 10 and the artificial intelligence server 20. If a communication error occurs, in S3, the artificial intelligence server 20 requests the surveillance camera 10 to transfer the video file again within a first period (every minute in FIG. 2) after the communication error occurs, and monitors again whether a communication error occurs between the surveillance camera 10 and the artificial intelligence server 20. On the other hand, in S4, when the transfer of the video file to the video storage unit 201a is completed, the artificial intelligence server 20 starts analyzing the video file recorded in the video storage unit 201a. Then, the artificial intelligence server 20 repeats the above S1 of requesting the surveillance camera 10 to transfer the video file to the video storage unit 201a during the second period (every 15 minutes in FIG. 2).
[0019] FIG. 3 shows a process of communicating several snapshot images (image files) between the surveillance camera 10 and the artificial intelligence server 20 in snapshot analysis. As an example, in FIG. 3, snapshot images taken by the surveillance camera 10 every minute are recorded in the recording medium 102. First, in S5, the artificial intelligence server 20 requests the surveillance camera 10 to transfer 15 minutes' worth of image files (corresponding to, for example, 15 snapshot images taken in a one-minute cycle that is the first cycle) to the image storage unit 201b. Next, in S6, the surveillance camera 10 transfers the 15 minutes' worth of image files to the image storage unit 201b. Then, the artificial intelligence server 20 starts analyzing the 15 minutes' worth of image files recorded in the image storage unit 201b.
[0020] The process of snapshot analysis in Fig. 3 can be executed in parallel with the process of video analysis in Fig. 2. Here, when both processes are executed in parallel, the analysis time can be shortened because the analysis is image analysis rather than video analysis. On the other hand, the snapshot analysis has a difference in the number of analysts compared to the video analysis.
[0021] With reference to FIG. 4, the overall processing flow of video analysis in the attribute analysis system 1 will be described. First, the artificial intelligence server 20 starts video analysis in step 400. Then, in step 410, the artificial intelligence server 20 requests the surveillance camera 10 to transfer the video recorded in the recording medium 102 to the video storage unit 201a. Then, the artificial intelligence server 20 monitors whether a communication error occurs between the surveillance camera 10 and the artificial intelligence server 20 in cooperation with the video analysis unit 202d of the central processing unit 202. In step 420, the surveillance camera 10 reads the video requested by the artificial intelligence server 20 from the recording medium 102, and transfers it to the video storage unit 201a of the artificial intelligence server 20 via the Internet line by the communication unit 103. Here, the communication unit 103 can transfer the video wirelessly, for example, by a SIM card built into the surveillance camera 10. In step 430, the artificial intelligence server 20 judges whether the transfer of the video requested in step 410 to the video storage unit 201a has been completed. Here, if it is determined in step 430 that the transfer of the video has been completed (successful), the process proceeds to step 440, where an artificial intelligence analysis is performed on the requested video. In this artificial intelligence analysis, the process cooperates with the moving object detection unit 202a and the artificial intelligence determination unit 202b of the central processing unit 202 to detect people from the video, cooperates with the marking unit 202c of the central processing unit 202 to mark the detected people (marking process) and display their attributes, and cooperates with the video analysis unit 202d of the central processing unit 202 to record the number and attributes of the detected people in the analysis data recording unit 201c. Then, in step 450, the video analysis is terminated. On the other hand, if it is determined in step 430 that the transfer is not complete, the process returns to step 410, where the artificial intelligence server 20 again requests the surveillance camera 10 to transfer the video recorded on the recording medium 102 to the video storage unit 201a. Here, the artificial intelligence server 20 can determine, in cooperation with the video analysis unit 202d of the central processing unit 202, whether the transfer of the requested video to the video storage unit 201a has been completed within a predetermined period (for example, every minute as a first period) after detecting the occurrence of a communication error. After the video analysis is completed, the artificial intelligence server 20 can return to step 400 at predetermined intervals (for example, every 15 minutes as the second interval) and start the video analysis again.
[0022] With reference to FIG. 5, the entire process flow of snapshot analysis in the attribute analysis system 1 will be described. First, the artificial intelligence server 20 starts snapshot analysis in step 500. Then, in step 510, the artificial intelligence server 20 requests the surveillance camera 10 to transfer images captured at a predetermined period (for example, one-minute period) from the recording medium 102 on which the images are recorded to the image storage unit 201b. In step 520, the surveillance camera 10 reads the images requested by the artificial intelligence server 20 from the recording medium 102 and transfers them to the image storage unit 201b of the artificial intelligence server 20 via the Internet line by the communication unit 103. Here, the communication unit 103 can transfer images wirelessly, for example, by a SIM card built into the surveillance camera 10. In step 530, the artificial intelligence server 20 judges whether the transfer of the images requested in step 510 to the image storage unit 201b has been completed. Here, if it is determined in step 530 that the image transfer is completed (successful), the process proceeds to step 540, where an artificial intelligence analysis is performed on the requested image. In this artificial intelligence analysis, the artificial intelligence server 20 cooperates with the artificial intelligence determination unit 202b of the central processing unit 202 to execute a process of detecting a person from the image, cooperates with the marking unit 202c of the central processing unit 202 to execute a process of marking the detected person (marking process), and cooperates with the snapshot analysis unit 202e of the central processing unit 202 to execute a process of recording the number of detected people in the analysis data recording unit 201c. Then, in step 550, the artificial intelligence server 20 applies a correction value determined in advance to the number of people detected by snapshot analysis to obtain a corrected number of snapshot analyzers, and records the corrected number of snapshot analyzers in the analysis data recording unit 201c. This correction value is a numerical value found in advance by performing a regression analysis on the number of people detected by video analysis of videos and images taken at the same time by the same surveillance camera 10 multiple times over a predetermined period (e.g., two months) and the number of people detected by snapshot analysis of the images (note that the details of the number-of-people correction in step 550 will be described later). Then, in step 560, the snapshot analysis ends. On the other hand, if it is determined in step 530 that the transfer has not been completed, the process proceeds to step 560 and the snapshot analysis ends. After the snapshot analysis is completed in step 560, the artificial intelligence server 20 can return to step 500 and start the snapshot analysis again at predetermined intervals (for example, every 15 minutes).
[0023] Next, referring to Figs. 6 to 9, snapshot analysis in which the accuracy of the number of people analyzed is improved by applying the correction value obtained from the video analysis will be described. Fig. 6 shows measurement data that compiles the number of people analyzing videos obtained from video analysis and the number of people analyzing snapshots (before correction) obtained from image analysis for each of video files and image files taken at the same time (having the same time stamp) by a surveillance camera 10 (having the same camera ID "1") installed at a specified location multiple times over a two-month period from June 1, 2023 to July 31, 2023. Referring to Fig. 6, it can be seen that there is a significant difference between the number of people analyzing snapshots (before correction) and the number of people analyzing videos.
[0024] Next, referring to FIG. 7, a method of performing regression analysis in the artificial intelligence server 20 to obtain an optimal correction value for the measurement data of the number of people who analyzed videos and the number of people who analyzed snapshots (before correction) shown in FIG. 6 will be described. Here, FIG. 7 shows a graph in which the vertical axis (y-axis) is the number of people who analyzed videos and the horizontal axis (x-axis) is the number of people who analyzed snapshots, plotting the measurement data of FIG. 6. Then, the artificial intelligence server 20 obtains an optimal correction value between the number of people who analyzed videos and the number of people who analyzed snapshots by linear approximation using, for example, the least squares method. As shown in FIG. 7, it can be seen that there is a relationship of y=1.8038x between the number of people who analyzed videos (y-axis) and the number of people who analyzed snapshots (x-axis) shown in FIG. 6. The above correction values can be calculated, for example, by another computer that communicates with the artificial intelligence server 20 receiving the above measurement data. Also, the calculated correction values can be transmitted to the artificial intelligence server 20 from the other computer.
[0025] FIG. 8 shows a table of basic statistics in the measurement data in FIG. 7. Referring to FIG. 8, the basic statistics in video analysis (corresponding to "video analysis" in FIG. 8) are a minimum of 874, a maximum of 3084, a median of 1415, an average of 1415, and a standard deviation of 2210. The basic statistics in snapshot analysis (corresponding to "snapshot analysis" in FIG. 8) are a minimum of 30, a maximum of 1748, a median of 769, an average of 820, and a standard deviation of 1718. And the error between the number of people analyzed by video analysis and the number of people analyzed by snapshot analysis at the same time (corresponding to "error (video-snapshot)" in FIG. 8) is a minimum of 234, a maximum of 2156, a median of 665, an average of 704, and a standard deviation of 241. Thus, if snapshot analysis is applied as it is without correction as a substitute for video analysis, the error with the number of people obtained by video analysis will be large.
[0026] FIG. 9 shows a graph of the error (before correction) of the number of snapshot analyzers (before correction) compared with the number of video analyzers, and the error (after correction) of the number of snapshot analyzers corrected using the correction value obtained from the regression analysis in FIG. 7. Here, in FIG. 9, the vertical axis is the error (number of people), and the horizontal axis is the number of samples (the number of pairs of the number of video analyzers and the number of snapshot analyzers taken at the same time). Referring to FIG. 9, the error (after correction) of the number of snapshot analyzers is closer to zero across the entire number of samples than the error (before correction) of the number of snapshot analyzers, and the difference with the number of video analyzers is smaller. In addition, the numerical values of the error (minimum value, maximum value, median value, average value, standard deviation) between the number of snapshot analyzers (after correction) and the number of video analyzers are shown in the "error result after correction" on the right of FIG. 9. In this "error result after correction," the median is 38.0 and the average value is 261.1. This value is smaller than the uncorrected value of snapshot analysis (i.e., the median of 665 and the average of 704 for "Error (video-snapshot)" in Figure 8). Therefore, by applying the correction value obtained from the regression analysis above to the snapshot analysis number of people, it is possible to obtain the number of people with an accuracy close to that of video analysis while reducing the time required compared to video analysis.
[0027] As described above, the attribute analysis system 1 according to one embodiment of the present invention has been described using an example in which the number of people detected in video analysis and snapshot analysis is calculated. However, it should be understood that the attribute analysis system 1 can be applied not only to the case of calculating the number of people, but also to the case of calculating the number of vehicles (e.g., traffic volume). For example, in the above embodiment, the analysis target can be vehicles, and the number and attributes of such vehicles (general cars, trucks, buses, motorcycles, etc.) can be analyzed. [Industrial Applicability]
[0028] The attribute analysis system, which is one embodiment of the present invention, can achieve highly reliable data communication between a surveillance camera and an artificial intelligence server, and is useful in that it can analyze the number of people with greater accuracy than analyzing images alone while reducing the time required by applying a correction value to the number of people analyzed in a snapshot. [Explanation of symbols]
[0029] 1. Attribute Analysis System 10. Surveillance Cameras 101 Power supply section 101a Battery 101b Solar Panel 102 Recording media (memory cards) 103 Communications Department 20 Artificial Intelligence Server 201 Records Department 201a Video storage section 201b Image storage section 201c Analysis data recording section 202 Central processing unit 202a Motion detection unit 202b Artificial Intelligence Judgment Department 202c Marking section 202d Video Analysis Department 202e Snapshot Analysis Section 30 Control console
Claims
1. a surveillance camera including a power supply unit including a battery and a solar panel for charging the battery, a recording medium for storing a captured video and an image captured in a first period, and a communication unit for transmitting the video or the image recorded in the recording medium; an artificial intelligence server including a video storage unit that records the video received from the recording medium, an image storage unit that records the image received from the recording medium, and a central processing unit that analyzes the video or the image to detect people and analyzes the number and attributes of the people; It is equipped with The artificial intelligence server, in the central processing unit, For the video and the images captured in advance multiple times at the same time during a predetermined period, the number of people detected from the video recorded in the video storage unit and the number of people detected from the images recorded in the image storage unit are calculated, and an approximation calculation is performed between the number of people detected from the video and the number of people detected from the images using a statistical method to calculate a correction value between the video and the images; The correction value is applied to the number of people detected in each of the images captured in the first period, to determine the number of people at the time the image was captured. An attribute analysis system comprising:
2. A surveillance camera including a power supply unit having a battery and a solar panel for charging the battery, a recording medium for storing captured video, and a communication unit for transmitting the video recorded on the recording medium; an artificial intelligence server including a video storage unit that records the video received from the recording medium, and a central processing unit that analyzes the video recorded in the video storage unit to detect people and analyzes the number and attributes of the people; It is equipped with The artificial intelligence server includes: (a) causing the surveillance camera to transmit video images captured over a predetermined period of time to the video storage unit, and monitoring for occurrence of a communication error between the surveillance camera and the artificial intelligence server; (b) when the video is transmitted to the video storage unit without detecting the communication error, the central processing unit is caused to analyze the video to detect people and to analyze the number and attributes of the people; (c) when the communication error is detected, causing the surveillance camera to transmit the video to the video storage unit again, and monitoring the occurrence of the communication error between the surveillance camera and the artificial intelligence server during a first period; (d) repeating the steps (a) to (c) for a second period longer than the first period. An attribute analysis system comprising:
3. The attribute analysis system according to claim 1 or 2, wherein the communication unit comprises a SIM card.
4. The attribute analysis system according to claim 1 or 2, further comprising an operation console that communicates with the artificial intelligence server via wired or wireless communication using an Internet line.
5. a surveillance camera including a power supply unit including a battery and a solar panel for charging the battery, a recording medium for storing a captured video and an image captured in a first period, and a communication unit for transmitting the video or the image recorded in the recording medium; an artificial intelligence server including a video storage unit that records the video received from the recording medium, an image storage unit that records the image received from the recording medium, and a central processing unit that analyzes the video or the image to detect people and analyzes the number and attributes of the people; An estimation method in an attribute analysis system comprising: The artificial intelligence server, in the central processing unit, a step of determining the number of people detected from the video recorded in the video storage unit and the number of people detected from the images recorded in the image storage unit for the video and the images taken multiple times at the same time during a predetermined period in advance, and performing an approximation calculation between the number of people detected from the video and the number of people detected from the images using a statistical method to determine a correction value between the video and the images; applying the correction value to the number of people detected in each of the images captured in the first period, to obtain the number of people at the time the image was captured; An estimation method in an attribute analysis system, comprising:
6. A surveillance camera including a power supply unit including a battery and a solar panel for charging the battery, a recording medium for storing a captured video, and a communication unit for transmitting the video recorded on the recording medium; an artificial intelligence server including a video storage unit that records the video received from the recording medium, and a central processing unit that analyzes the video recorded in the video storage unit to detect people and analyzes the number and attributes of the people; A communication method in an attribute analysis system comprising: The artificial intelligence server includes: (a') causing the surveillance camera to transmit video images captured over a predetermined period of time to the video storage unit, and monitoring for occurrence of a communication error between the surveillance camera and the artificial intelligence server; (b') when the video is transmitted to the video storage unit without detecting the communication error, causing the central processing unit to analyze the video to detect people and analyze the number and attributes of the people; (c') when the communication error is detected, causing the surveillance camera to transmit the video to the video storage unit again, and monitoring the occurrence of the communication error between the surveillance camera and the artificial intelligence server during a first period; (d') repeating the steps (a') to (c') above during a second period longer than the first period; A communication method in an attribute analysis system.
Citation Information
Patent Citations
Pedestrian counting method based on group context
CN102289817A
Qoe analysis-based video frame management method and apparatus
CN107454446A
Image prediction method and device, electronic equipment and storage medium
CN111008606A
Information offer system for real-time data transmission
JP1997182055A
Apparatus and method for processing moving image as well as recording medium
JP2002077798A
