A campus safety risk monitoring method and system

Abstract

The present invention discloses a campus safety risk monitoring method and system, which relate to the field of campus safety technology. It solves the problems that existing campus safety monitoring cannot discover potential risks in combination with other data; cannot timely warn and prompt standardized processing procedures; and cannot timely detect public opinion information. The solution includes: obtaining monitoring data and psychological assessment data; determining key personnel based on the psychological assessment data; identifying the risk level of personnel behavior based on the monitoring data, and generating warning signals based on graded warning rules. The graded warning rules include: generating warning signals for medium-risk behaviors or all high-risk behaviors of key personnel; pushing warning signals to corresponding responsible management personnel and nearby patrol personnel; pushing related guidance plans based on the type of accident feedback from the nearby patrol personnel, and the related guidance plans are used to guide the standardized handling of accidents; generating keywords based on the type of accident to monitor public opinion on the website; and solving the above problems.

Description

Technical Field

[0001] The present invention relates to the field of campus safety technology, and more specifically, to a campus safety risk monitoring method and system. Background Art

[0002] Current campus security monitoring technology solutions primarily encompass video surveillance systems, IoT (Internet of Things) technology, and AI-powered behavioral analysis. Video surveillance systems deploy high-definition cameras to cover key areas of a school and employ intelligent monitoring algorithms to automatically identify abnormal behavior. IoT technology, through sensors, cameras, and other devices, enables real-time monitoring, early warning, and management of campus security conditions.

[0003] However, the above-mentioned campus safety monitoring technology solutions all exist in isolation. They cannot be combined with other data to discover potential risks before an accident occurs; they cannot provide timely warnings and prompt standardized handling procedures when an accident occurs; and they cannot detect and process public opinion information in a timely manner after an accident occurs.

[0004] Therefore, the present application provides a campus safety risk monitoring method and system to solve the above problems. Summary of the Invention

[0005] The purpose of this application is to provide a campus safety risk monitoring method and system, which solves the problems of existing campus safety monitoring that is unable to combine other data to discover potential risks before an accident occurs; unable to timely warn and prompt standardized processing procedures when an accident occurs; and unable to timely detect and process public opinion information after an accident occurs. This application combines psychological assessment data before an accident occurs to discover potential risk personnel and pays special attention to them in subsequent risk behavior monitoring; provides timely warnings to responsible management personnel and nearby patrol personnel when an accident occurs, and provides standardized processing procedures; and promptly monitors public opinion information after an accident occurs to avoid the situation from worsening.

[0006] This application first provides a campus safety risk monitoring method, including: obtaining multi-dimensional monitoring data, the multi-dimensional monitoring data including: monitoring data and psychological assessment data; determining key personnel based on the psychological assessment data; identifying the risk level of personnel behavior based on the monitoring data, the risk levels including: low risk, medium risk and high risk, generating warning signals based on graded warning rules, the graded warning rules including: generating warning signals for medium-risk behaviors or all high-risk behaviors of key personnel; pushing warning signals to corresponding responsible management personnel and nearby patrol personnel; pushing related guidance plans based on the type of accident reported by the nearby patrol personnel, the related guidance plans are used to guide the standardized handling of accidents; generating keywords based on the type of accident to monitor public opinion on the website.

[0007] In one possible implementation, the risk level of human behavior is identified based on monitoring data, and the risk levels include: low risk, medium risk and high risk. A warning signal is generated based on a hierarchical warning rule, and the hierarchical warning rule includes: generating a warning signal for medium-risk behavior of key personnel or all high-risk behaviors; including: cutting the monitoring data into ordered monitoring images; inputting multiple frames of monitoring images into a YOLOv8-CNN-LSTM model according to the detection frequency to obtain the risk level, and the risk levels include: low risk, medium risk and high risk; when medium-risk human behavior is identified in multiple frames of monitoring images, face recognition is performed on the multiple frames of monitoring images and the number of times key personnel appear in the multiple frames of monitoring images is greater than the average number of times other non-aggregated personnel appear, and if so, a warning signal is generated; when high-risk human behavior is identified in multiple frames of monitoring images, a warning signal is directly generated, and the warning signal includes: multiple frames of monitoring images, monitoring equipment location information and risk level.

[0008] In one possible implementation, the YOLOv8-CNN-LSTM model includes: sequentially inputting multiple frames of surveillance footage into a YOLOv8 network to obtain object detection frames and object detection frame diagrams with corresponding IDs, where the object detection frames include human body frames and human skeleton frames; inputting the object detection frame diagrams into a CNN network to extract feature vectors; inputting the feature vectors into an LSTM network to obtain time series feature vectors; and classifying the time series feature vectors through a fully connected layer, where the classifications include low risk, medium risk, and high risk.

[0009] In one possible implementation, when a medium-risk behavior of a person is identified in a multi-frame surveillance image, face recognition is performed on the multi-frame surveillance image and the number of times the key focus person appears in the multi-frame surveillance image is counted to see whether it is greater than the average number of times other non-aggregated persons appear. If it is greater, an early warning signal is generated; including: when a medium-risk behavior of a person is identified in a multi-frame surveillance image, an object detection frame diagram of the multi-frame surveillance image is obtained, face recognition is performed on each ID object detection frame in the object detection frame diagram, and when a key focus person is identified, the corresponding ID is marked as the target ID, and multiple object detection frames are counted. The number of occurrences of the target ID in the image; for non-target IDs, calculate the distance between the non-target ID object detection frames in each object detection frame map, and calculate the average distance. The non-target IDs corresponding to the object detection frames with distances less than the average distance in multiple object detection frame maps are marked as clustered IDs, and the other non-target IDs are marked as non-clustered IDs; count the number of occurrences of non-clustered IDs in multiple object detection frame maps, and calculate the average number of occurrences of each non-clustered ID as the medium-risk warning threshold; compare the number of occurrences of the target ID in multiple object detection frame maps with the medium-risk warning threshold, and generate a warning signal if it is greater than the medium-risk warning threshold.

[0010] In one possible implementation, the early warning signal is pushed to the corresponding responsible management personnel and the nearest patrol personnel; including: obtaining the location information of the monitoring equipment according to the early warning signal, obtaining the positioning information of each patrol personnel, calculating the distance between the monitoring equipment and each patrol personnel according to the address of the monitoring equipment and the positioning information of each patrol personnel, determining the nearest patrol personnel of the monitoring equipment, and sending an early warning signal to the nearest patrol personnel; determining the responsible management personnel of the monitoring equipment according to the location information of the monitoring equipment and the scope of responsibility of each management personnel, and sending an early warning signal to the responsible management personnel.

[0011] In one possible implementation, search keywords are generated based on the accident type to monitor public opinion on the website; this includes: generating multiple event keywords based on the accident type, and generating multiple location keywords based on the school name; obtaining multiple search keywords based on the Cartesian product of the event keywords and the location keywords; crawling relevant information on the monitoring website based on the multiple search keywords to obtain public opinion monitoring results.

[0012] The present application also provides a campus safety risk monitoring system for implementing a campus safety risk monitoring method as described above; it includes: a data acquisition unit for acquiring multi-dimensional monitoring data, the multi-dimensional monitoring data including: monitoring data and psychological assessment data; a personnel determination unit for determining key personnel based on the psychological assessment data; an early warning generation unit for identifying the risk level of personnel behavior based on the monitoring data, the risk levels including: low risk, medium risk and high risk, and generating early warning signals based on graded early warning rules, the graded early warning rules including: generating early warning signals for medium-risk behaviors or all high-risk behaviors of key personnel; an early warning sending unit for pushing early warning signals to corresponding responsible management personnel and nearby patrol personnel; a guidance plan unit for pushing associated guidance plans based on the type of accident reported by the nearby patrol personnel, the associated guidance plans are used to guide the standardized handling of accidents; a public opinion monitoring unit for generating search keywords based on the type of accident to monitor public opinion on the website.

[0013] In one possible embodiment, the early warning generation unit specifically includes: a video processing unit, which is used to cut the monitoring data into ordered monitoring pictures; a risk prediction unit, which is used to input multiple frames of monitoring pictures into the YOLOv8-CNN-LSTM model according to the detection frequency to obtain a risk level, and the risk levels include: low risk, medium risk and high risk; a graded early warning unit, which is used to perform face recognition on the multiple frames of monitoring pictures and count whether the number of times the key focus persons appear in the multiple frames of monitoring pictures is greater than the average number of times other non-aggregated persons appear in the multiple frames of monitoring pictures when medium-risk human behavior is identified in the multiple frames of monitoring pictures, and generate an early warning signal if so; when high-risk human behavior is identified in the multiple frames of monitoring pictures, a early warning signal is directly generated, and the early warning signal includes: multiple frames of monitoring pictures, monitoring equipment location information and risk level.

[0014] In one possible implementation, the YOLOv8-CNN-LSTM model in the risk prediction unit is specifically used to: input multiple frames of surveillance images into the YOLOv8 network in sequence to obtain object detection frames and object detection frame diagrams with corresponding IDs, where the object detection frames include human body frames and human skeleton frames; input the object detection frame diagrams into the CNN network to extract feature vectors; input the feature vectors into the LSTM network to obtain time series feature vectors; and classify the time series feature vectors through a fully connected layer, where the classifications include low risk, medium risk, and high risk.

[0015] In one possible embodiment, the graded warning unit is specifically used to: when medium-risk human behavior is identified in multiple frames of monitoring images, obtain the object detection frame diagram of the multiple frames of monitoring images, perform face recognition on each ID object detection frame in the object detection frame diagram, and when a key person is identified, mark the corresponding ID as the target ID, and count the number of occurrences of the target ID in multiple object detection frame diagrams; for non-target IDs, calculate the distance between the non-target ID object detection frames in each object detection frame diagram, and calculate the average distance, mark the non-target IDs corresponding to the object detection frames in multiple object detection frame diagrams whose distances are less than the average distance as clustered IDs, and mark other non-target IDs as non-clustered IDs; count the number of occurrences of non-clustered IDs in multiple object detection frame diagrams, and calculate the average number of occurrences of each non-clustered ID as a medium-risk warning threshold; compare the number of occurrences of the target ID in multiple object detection frame diagrams with the medium-risk warning threshold, and generate a warning signal if it is greater than the medium-risk warning threshold.

[0016] Compared with the existing technology, the present application has the following beneficial effects: it proposes risk monitoring linked with psychological assessment data, which can screen out students with serious psychological symptoms in advance and focus on them; it adopts graded warning to warn all high-risk behaviors, and also warn the medium-risk behaviors of key personnel to avoid potential accidents; it adopts a model composed of YOLOv8, CNN, LSTM, and fully connected layers in sequence to effectively identify and classify dynamic behaviors in videos; it adopts a dual warning mode, on the one hand, sending warning signals to nearby patrol personnel, and on the other hand, sending warning signals to responsible management personnel, so as to facilitate simultaneous online and offline processing; it conducts public opinion monitoring after the accident, so as to promptly guide the positive development of public opinion and avoid the negative impact caused by untimely processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0018] Figure 1A flowchart of a campus safety risk monitoring method provided in an embodiment of the present application;

[0019] Figure 2 This is a structural diagram of the campus safety risk monitoring system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] Hereinafter, the terms "include" or "may include" as used in various embodiments of the present application indicate the presence of the claimed function, operation, or element, and do not limit the addition of one or more functions, operations, or elements. In addition, as used in various embodiments of the present application, the terms "include", "have" and their cognates are intended only to indicate specific features, numbers, steps, operations, elements, components, or combinations of the foregoing, and should not be understood as first excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing or the possibility of adding one or more features, numbers, steps, operations, elements, components, or combinations of the foregoing.

[0021] The terms used in the various embodiments of the present application are only used for the purpose of describing specific embodiments and are not intended to limit the various embodiments of the present application. As used herein, the singular form is intended to also include the plural form, unless the context clearly indicates otherwise. Unless otherwise specified, all terms used herein (including technical terms and scientific terms) have the same meaning as those commonly understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meaning as the contextual meaning in the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning, unless clearly defined in the various embodiments of the present application.

[0022] In order to make the objectives, technical solutions and advantages of this application more clear, the present application is further described in detail below in conjunction with examples and drawings. The schematic implementation methods of this application and their descriptions are only used to explain this application and are not intended to limit this application.

[0023] See Figure 1 As shown, Figure 1A flowchart of a campus safety risk monitoring method provided in an embodiment of the present application. The method includes: S1, obtaining multi-dimensional monitoring data, the multi-dimensional monitoring data including: monitoring data and psychological assessment data; S2, determining key personnel based on the psychological assessment data; S3, identifying the risk level of personnel behavior based on the monitoring data, the risk level including: low risk, medium risk, and high risk, and generating a warning signal based on a graded warning rule, the graded warning rule including: generating a warning signal for medium-risk behavior or all high-risk behaviors of key personnel; S4, pushing the warning signal to the corresponding responsible management personnel and nearby patrol personnel; S5, pushing the associated guidance plan based on the accident type reported by the nearby patrol personnel, the associated guidance plan is used to guide the standardized handling of the accident; S6, generating keywords based on the accident type to monitor public opinion on the website.

[0024] Specifically, monitoring data can come from the campus's video surveillance system, and psychological assessment data can come from psychological assessment questionnaires filled out by students. Psychological assessment data is used to assess the mental health of individuals, and individuals with severe psychological symptoms are identified as key personnel. Monitoring data is analyzed to identify low-risk, medium-risk, and high-risk behaviors, and selective warnings are issued for medium and high risks based on graded warning rules. Warning signals are pushed to responsible management personnel and nearby patrol personnel for simultaneous online and offline supervision. Based on the type of accident reported by the nearby patrol personnel, relevant guidance plans are pushed to help patrol personnel handle accidents in a standardized manner. Finally, search keywords are generated based on the type of accident to monitor public opinion on the website and promptly understand the public's reaction and emotions to the accident.

[0025] The improvement of this application lies in combining psychological assessment data to identify potential risk persons before an accident occurs and giving them special attention in subsequent risk behavior monitoring; providing timely warnings to responsible management personnel and nearby patrol personnel when an accident occurs and providing standardized handling procedures; and monitoring public opinion information in a timely manner after an accident occurs to avoid the situation from worsening.

[0026] Step S1 involves acquiring multi-dimensional monitoring data, which includes monitoring data and psychological assessment data. Specifically, monitoring data typically refers to data collected by various campus monitoring devices (such as cameras and sensors), used to monitor and record information such as a subject's behavior and status in real time. Psychological assessment data refers to data collected through psychological assessment tools, questionnaires, and interviews, used to assess an individual's psychological state, emotions, and mental health. By combining monitoring data with psychological assessment data, individuals with potential risks can be identified and prioritized.

[0027] Step S2 involves identifying individuals of key concern based on psychological assessment data. Specifically, psychological assessment data is collected from students, such as the National Middle School Student Mental Health Scale (MSSMHS). Scores are calculated based on the scale, such as for depression, anxiety, hostility, and paranoia. Individuals with severe psychological symptoms, such as severe depression or severe anxiety, are identified as individuals of key concern based on the range of their scores.

[0028] It is understandable that, unlike traditional campus risk monitoring, this application proposes risk monitoring linked with psychological assessment data, which can screen out students with serious psychological symptoms in advance and focus on them.

[0029] Step S3 is to identify the risk level of human behavior based on the monitoring data. The risk levels include: low risk, medium risk, and high risk. A warning signal is generated based on the graded warning rules. The graded warning rules include: generating a warning signal for medium-risk behavior of key personnel or all high-risk behaviors. In one possible implementation, S3 includes: S31, cutting the monitoring data into ordered monitoring images; S32, inputting multiple frames of monitoring images into the YOLOv8-CNN-LSTM model based on the detection frequency to obtain the risk level. The risk levels include: low risk, medium risk, and high risk; S33, when medium-risk human behavior is identified in the multiple frames of monitoring images, performing face recognition on the multiple frames of monitoring images and counting whether the number of times the key personnel appear in the multiple frames of monitoring images is greater than the average number of times other non-aggregated personnel appear. If so, a warning signal is generated; when high-risk human behavior is identified in the multiple frames of monitoring images, a warning signal is directly generated. The warning signal includes: multiple frames of monitoring images, monitoring device location information, and risk level.

[0030] Specifically, the surveillance data is in video format. The video is first broken down into ordered frames. Each frame undergoes preprocessing, including resizing, normalization, and transposition. Multiple frames are then fed into the YOLOv8-CNN-LSTM model based on detection frequency. For example, a risk level prediction is performed for every eight frames. The model outputs a predicted risk level: low, medium, or high. For example, medium risk can include situations such as lingering near railings and gatherings; high risk can include situations such as climbing over railings and engaging in physical fights; and low risk can include other situations excluding the aforementioned two categories. Next, based on hierarchical warning rules, warnings are issued for medium-risk behaviors performed by key individuals and for all high-risk behaviors. A warning signal is generated, including the multiple frames of surveillance footage where the risky behavior was detected, the device address of the monitoring device to which the current monitoring data belongs, and the model's predicted risk level.

[0031] In addition, in this application, instead of issuing warnings for all medium-risk behaviors in which key persons of concern appear, it is proposed to compare the average number of appearances of non-aggregated persons in multi-frame monitoring images with the number of appearances of key persons of concern in multi-frame monitoring images to determine whether the key persons of concern may be involved in the medium-risk behavior and reduce false alarms. The reason for using the average number of appearances of non-aggregated persons in multi-frame monitoring images as the medium-risk warning threshold is that when the number of appearances is used as the basis for judgment, the number of appearances of aggregated persons has a greater impact on the judgment, and most risky behaviors often have a tendency of people gathering. Therefore, this application proposes to exclude the number of appearances of aggregated persons and use the number of appearances of non-aggregated persons as the normal situation. When the number of appearances of key persons of concern is greater than the number of appearances of non-aggregated persons, it means that the key persons of concern are very likely to be involved in the medium-risk behavior, thereby generating a warning signal for the medium-risk behavior.

[0032] It is understandable that, unlike traditional campus risk monitoring solutions, this application uses a graded warning system, issuing warnings for all high-risk behaviors, while also issuing warnings for medium-risk behaviors involving key personnel to avoid potential accidents. For example, when a key personnel is found to be in a crowd for a long time, a warning will be given to prevent them from being harmed. At the same time, this application only performs facial recognition on monitoring footage of medium-risk behaviors, reducing the amount of data calculations while ensuring the safety of key personnel.

[0033] Furthermore, the YOLOv8-CNN-LSTM model includes: inputting multiple frames of surveillance images into the YOLOv8 network in sequence to obtain object detection frames and object detection frame diagrams with corresponding IDs, where the object detection frames include human body frames and human skeleton frames; inputting the object detection frame diagrams into the CNN network to extract feature vectors; inputting the feature vectors into the LSTM network to obtain time series feature vectors; and classifying the time series feature vectors through a fully connected layer, where the classifications include low risk, medium risk, and high risk.

[0034] Specifically, the front end of the YOLOv8-CNN-LSTM model is the YOLO v8 network, which is used to detect and draw the human body and skeleton in each frame of the surveillance image to obtain a detection frame diagram, which provides assistance for subsequent risk behavior identification; the middle end of the model is the CNN-LSTM network, CNN is used to extract feature vectors from the detection frame diagram, and LSTM is used to obtain time series feature vectors based on the feature vectors that change over time. The time series feature vectors can represent temporal behaviors, such as: people staying, people gathering, people climbing over railings, etc.; the back end of the model is a fully connected layer, which is used to classify time series feature vectors into low risk, medium risk or high risk.

[0035] The YOLOv8-CNN-LSTM model is trained using the following method: video data labeled with risk levels is obtained; the video data is augmented by flipping, shrinking, and enlarging the data; 70% of the augmented video data is used as training data and 30% as test data. The YOLOv8-CNN-LSTM model is trained using the training data and tested using the test data to obtain a trained YOLOv8-CNN-LSTM model.

[0036] It can be understood that the risk behavior identification of this application adopts a model composed of YOLOv8, CNN, LSTM, and fully connected layers in sequence. It performs target detection based on YOLOv8, extracts the features of each frame based on CNN, captures time series features based on LSTM, and finally performs classification through the fully connected layer, which can effectively identify and classify dynamic behaviors in videos.

[0037] Furthermore, when it is identified that there is a medium-risk behavior of a person in multiple frames of monitoring images, face recognition is performed on the multiple frames of monitoring images and the number of times the key focus person appears in the multiple frames of monitoring images is counted to see if it is greater than the average number of times other non-aggregated people appear. If it is greater, an early warning signal is generated; including: when it is identified that there is a medium-risk behavior of a person in multiple frames of monitoring images, an object detection frame diagram of the multiple frames of monitoring images is obtained, face recognition is performed on each ID object detection frame in the object detection frame diagram, and when a key focus person is identified, the corresponding ID is marked as the target ID, and the target ID in multiple object detection frame diagrams is counted. The number of occurrences of the ID; for non-target IDs, calculate the distance between the non-target ID object detection frames in each object detection frame map, and calculate the average distance. The non-target IDs corresponding to the object detection frames whose distances in multiple object detection frame maps are less than the average distance are marked as clustered IDs, and the other non-target IDs are marked as non-clustered IDs; count the number of occurrences of non-clustered IDs in multiple object detection frame maps, and calculate the average number of occurrences of each non-clustered ID as the medium-risk warning threshold; compare the number of occurrences of the target ID in multiple object detection frame maps with the medium-risk warning threshold, and generate a warning signal if it is greater than the medium-risk warning threshold.

[0038] It can be understood that this application uses YOLOv8 to detect targets and mark IDs in surveillance images, and quickly counts the number of appearances of key personnel and non-aggregated personnel based on ID and face recognition, thereby providing early warnings for medium-risk behaviors that key personnel may be involved in, while reducing false alarms.

[0039] Step S4 is to push the warning signal to the corresponding responsible management personnel and the nearest patrol personnel. In one possible implementation, S4 includes: obtaining the monitoring device location information and the location information of each patrol personnel based on the warning signal; calculating the distance between the monitoring device and each patrol personnel based on the monitoring device address and the location information of each patrol personnel; determining the nearest patrol personnel of the monitoring device; and sending the warning signal to the nearest patrol personnel; determining the responsible management personnel of the monitoring device based on the monitoring device location information and the responsibility scope of each management personnel; and sending the warning signal to the responsible management personnel.

[0040] Specifically, the location information of the monitoring device that captured the risky behavior is extracted from the warning signal. This location information is compared with the positioning information of each patrol officer to determine the patrol officer closest to the monitoring device. A warning signal is then sent to the nearest patrol officer, allowing them to rush to the scene and address the risky behavior in a timely manner. Simultaneously, the location information of the monitoring device is compared with the pre-set responsibility ranges of each administrator to determine the responsible manager for the monitoring device. A warning signal is then sent to the responsible manager, allowing them to call up the monitoring device's footage for online monitoring. Warning signals can be delivered to the corresponding patrol officers and managers via various push methods, such as SMS, APP, and phone calls.

[0041] It is understandable that this application adopts a dual warning mode, on the one hand sending warning signals to nearby patrol personnel, and on the other hand sending warning signals to responsible management personnel, so as to facilitate simultaneous online and offline processing.

[0042] Step S5 is to push the associated guidance plan based on the accident type reported by the nearest patrol officer. The associated guidance plan is used to guide the standardized handling of the accident. Specifically, guidance plans can be prepared in advance according to different accident types. After obtaining the accident type reported by the nearest patrol officer, the corresponding guidance plan is matched and sent to the patrol officer. This allows for standardized handling of the accident and avoids omissions.

[0043] Step S6 generates search keywords based on the accident type and monitors public opinion on the website. In one possible implementation, S6 includes: generating multiple event keywords based on the accident type and multiple location keywords based on the school name; obtaining multiple search keywords based on the Cartesian product of the event keywords and location keywords; and crawling relevant information on the monitoring website based on the multiple search keywords to obtain public opinion monitoring results.

[0044] Specifically, several commonly used social networking sites can be set up as public opinion monitoring sites. Multiple keywords can be expanded based on the type of accident. For example, synonyms and related words for the accident type can be generated using the existing synonym library API as event keywords; location keywords can be generated based on the school and its abbreviation. Event keywords and location keywords can be combined using a Cartesian product to generate multiple search keywords. Based on the search keywords, relevant information can be crawled from the monitoring site and stored in documents for display, facilitating timely monitoring of public opinion trends after an accident occurs.

[0045] It is understandable that, unlike traditional campus risk monitoring programs, this application conducts public opinion monitoring after an accident occurs, so as to promptly guide public opinion in a positive direction and avoid negative impacts caused by untimely handling.

[0046] Please attend Figure 2 As shown, Figure 2 A structural diagram of a campus safety risk monitoring system provided in an embodiment of the present application. The system is used to implement the above-mentioned campus safety risk monitoring method, and the system includes: a data acquisition unit for acquiring multi-dimensional monitoring data, the multi-dimensional monitoring data including: monitoring data and psychological assessment data; a personnel determination unit for determining key personnel based on the psychological assessment data; an early warning generation unit for identifying the risk level of personnel behavior based on the monitoring data, the risk level including: low risk, medium risk and high risk, and generating early warning signals based on hierarchical early warning rules, the hierarchical early warning rules including: generating early warning signals for medium-risk behaviors or all high-risk behaviors of key personnel; an early warning sending unit for pushing early warning signals to corresponding responsible management personnel and nearby patrol personnel; a guidance plan unit for pushing associated guidance plans based on the type of accident reported by the nearby patrol personnel, the associated guidance plans being used to guide the standardized handling of accidents; and a public opinion monitoring unit for generating search keywords based on the type of accident to monitor public opinion on the website.

[0047] In one possible embodiment, the early warning generation unit specifically includes: a video processing unit, which is used to cut the monitoring data into ordered monitoring pictures; a risk prediction unit, which is used to input multiple frames of monitoring pictures into the YOLOv8-CNN-LSTM model according to the detection frequency to obtain a risk level, and the risk levels include: low risk, medium risk and high risk; a graded early warning unit, which is used to perform face recognition on the multiple frames of monitoring pictures and count whether the number of times the key focus persons appear in the multiple frames of monitoring pictures is greater than the average number of times other non-aggregated persons appear in the multiple frames of monitoring pictures when medium-risk human behavior is identified in the multiple frames of monitoring pictures, and generate an early warning signal if so; when high-risk human behavior is identified in the multiple frames of monitoring pictures, a early warning signal is directly generated, and the early warning signal includes: multiple frames of monitoring pictures, monitoring equipment location information and risk level.

[0048] Furthermore, the YOLOv8-CNN-LSTM model in the risk prediction unit is specifically used to: input multiple frames of surveillance images into the YOLOv8 network in sequence to obtain object detection frames and object detection frame diagrams with labeled object detection frames and corresponding IDs, where the object detection frames include human body frames and human skeleton frames; input the object detection frame diagrams into the CNN network to extract feature vectors; input the feature vectors into the LSTM network to obtain time series feature vectors; and classify the time series feature vectors through a fully connected layer, where the classifications include low risk, medium risk, and high risk.

[0049] Furthermore, the graded warning unit is specifically used to: when medium-risk human behavior is identified in multiple frames of monitoring images, obtain the object detection frame diagram of the multiple frames of monitoring images, perform face recognition on each ID object detection frame in the object detection frame diagram, and when a key person is identified, mark the corresponding ID as the target ID, and count the number of occurrences of the target ID in multiple object detection frame diagrams; for non-target IDs, calculate the distance between the non-target ID object detection frames in each object detection frame diagram, and calculate the average distance, mark the non-target IDs corresponding to the object detection frames in multiple object detection frame diagrams whose distances are less than the average distance as clustered IDs, and mark other non-target IDs as non-clustered IDs; count the number of occurrences of non-clustered IDs in multiple object detection frame diagrams, and calculate the average number of occurrences of each non-clustered ID as the medium-risk warning threshold; compare the number of occurrences of the target ID in multiple object detection frame diagrams with the medium-risk warning threshold, and generate a warning signal if it is greater than the medium-risk warning threshold.

[0050] In one possible implementation, the early warning sending unit is specifically used to: obtain the location information of the monitoring device based on the early warning signal, obtain the positioning information of each patrol officer, calculate the distance between the monitoring device and each patrol officer based on the address of the monitoring device and the positioning information of each patrol officer, determine the nearest patrol officer of the monitoring device, and send an early warning signal to the nearest patrol officer; determine the responsible manager of the monitoring device based on the location information of the monitoring device and the scope of responsibility of each manager, and send an early warning signal to the responsible manager.

[0051] In one possible implementation, the public opinion monitoring unit specifically includes: a first keyword generation unit, used to generate multiple event keywords based on the accident type, and generate multiple location keywords based on the school name; a second keyword generation unit, used to obtain multiple search keywords based on the Cartesian product of the event keywords and the location keywords; a public opinion retrieval unit, used to crawl relevant information on the monitoring website based on the multiple search keywords to obtain public opinion monitoring results.

[0052] It can be understood that the campus safety risk monitoring system provided in the embodiment of the present application is used to implement the above-mentioned campus safety risk monitoring method, corresponds one-to-one with the above-mentioned method, and has corresponding technical effects, so it will not be elaborated on.

[0053] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A campus safety risk monitoring method, characterized in that: include: Acquiring multi-dimensional monitoring data, wherein the multi-dimensional monitoring data includes: monitoring data and psychological assessment data; Determining key personnel based on the psychological assessment data; Identify the risk level of personnel behavior based on the monitoring data, where the risk level includes low risk, medium risk, and high risk, and generate a warning signal based on a graded warning rule, where the graded warning rule includes generating a warning signal for medium-risk behavior or all high-risk behaviors of the key personnel; Push the warning signal to the corresponding responsible management personnel and nearby patrol personnel; Push related guidance plans based on the accident type reported by the nearest patrol personnel. The related guidance plans are used to guide the standardized handling of the accident. Generate search keywords based on the accident type to monitor public opinion on the website; Among them, the risk level of personnel behavior is identified according to the monitoring data, and the risk level includes: low risk, medium risk and high risk. A warning signal is generated based on a graded warning rule, and the graded warning rule includes: generating a warning signal for the medium-risk behavior of the key personnel or all high-risk behaviors; including: cutting the monitoring data into ordered monitoring pictures; inputting multiple frames of monitoring pictures into the YOLOv8-CNN-LSTM model according to the detection frequency to obtain the risk level, and the risk level includes: low risk, medium risk and high risk; when medium-risk personnel behavior is identified in multiple frames of monitoring pictures, face recognition is performed on the multiple frames of monitoring pictures and statistics are counted to see whether the number of key personnel appearing in the multiple frames of monitoring pictures is greater than the average number of other non-aggregated personnel. If it is greater, a warning signal is generated; when high-risk personnel behavior is identified in multiple frames of monitoring pictures, a warning signal is directly generated, and the warning signal includes: multiple frames of monitoring pictures, monitoring equipment location information and risk level; When it is identified that there is a medium-risk behavior of a person in a multi-frame monitoring picture, face recognition is performed on the multi-frame monitoring picture and the number of times the key focus person appears in the multi-frame monitoring picture is counted to see if it is greater than the average number of times other non-aggregated people appear. If it is greater, an early warning signal is generated; including: when it is identified that there is a medium-risk behavior of a person in a multi-frame monitoring picture, an object detection frame diagram of the multi-frame monitoring picture is obtained based on the YOLOv8 network, face recognition is performed on each ID object detection frame in the object detection frame diagram, and when a key focus person is identified, the corresponding ID is marked as the target ID, and multiple object detection frames are counted. The number of occurrences of the target ID in the frame map; for non-target IDs, the distance between the object detection frames of the non-target IDs in each object detection frame map is calculated, and the average distance is calculated. The non-target IDs corresponding to the object detection frames whose distances in multiple object detection frame maps are less than the average distance are marked as clustered IDs, and the other non-target IDs are marked as non-clustered IDs; the number of occurrences of non-clustered IDs in multiple object detection frame maps is counted, and the average number of occurrences of each non-clustered ID is calculated as the medium-risk warning threshold; the number of occurrences of the target ID in multiple object detection frame maps is compared with the medium-risk warning threshold, and a warning signal is generated if it is greater than the medium-risk warning threshold.

2. A campus safety risk monitoring method according to claim 1, characterized in that: The YOLOv8-CNN-LSTM model includes: Input multiple frames of surveillance images into the YOLOv8 network in sequence to obtain an object detection frame with annotated object detection frames and corresponding IDs. The object detection frames include: a human body frame and a human skeleton frame; Input the object detection block diagram into the CNN network to extract the feature vector; Input the feature vector into the LSTM network to obtain a time series feature vector; The time series feature vector is classified through a fully connected layer, and the classification includes: low risk, medium risk and high risk.

3. A campus safety risk monitoring method according to claim 1, characterized in that: Push the warning signal to the corresponding responsible management personnel and nearby patrol personnel; including: Obtaining the location information of the monitoring device according to the warning signal, obtaining the positioning information of each patrol officer, calculating the distance between the monitoring device and each patrol officer according to the address of the monitoring device and the positioning information of each patrol officer, determining the nearest patrol officer of the monitoring device, and sending a warning signal to the nearest patrol officer; The responsible manager of the monitoring equipment is determined according to the location information of the monitoring equipment and the scope of responsibility of each manager, and an early warning signal is sent to the responsible manager.

4. A campus safety risk monitoring method according to claim 1, characterized in that: Generate search keywords based on the accident type to monitor public opinion on the website; including: Generate multiple event keywords based on the accident type, and generate multiple location keywords based on the school name; Obtaining a plurality of search keywords according to the Cartesian product of the event keyword and the location keyword; Based on the multiple search keywords, relevant information is crawled on the monitoring website to obtain public opinion monitoring results.

5. A campus safety risk monitoring system, characterized in that: A campus safety risk monitoring method for implementing any one of claims 1 to 4, comprising: A data acquisition unit, configured to acquire multi-dimensional monitoring data, wherein the multi-dimensional monitoring data includes monitoring data and psychological assessment data; A personnel determination unit, configured to determine key personnel based on the psychological assessment data; an early warning generating unit, configured to identify a risk level of personnel behavior based on the monitoring data, the risk level including low risk, medium risk, and high risk, and generate an early warning signal based on a graded early warning rule, the graded early warning rule including generating an early warning signal for medium-risk behavior or all high-risk behaviors of the key personnel; An early warning sending unit is used to push the early warning signal to the corresponding responsible management personnel and nearby patrol personnel; a guidance plan unit is used to push the associated guidance plan based on the accident type reported by the nearby patrol personnel, and the associated guidance plan is used to guide the standardized handling of the accident; The public opinion monitoring unit is used to generate search keywords based on the accident type to monitor public opinion on the website.

6. A campus safety risk monitoring system according to claim 5, characterized in that: The warning generation unit specifically includes: A video processing unit, configured to cut the monitoring data into ordered monitoring images; A risk prediction unit is used to input multiple frames of monitoring images into the YOLOv8-CNN-LSTM model according to the detection frequency to obtain a risk level, wherein the risk level includes: low risk, medium risk and high risk; The graded warning unit is used to perform face recognition on the multiple frames of monitoring images and count whether the number of times the key focus persons appear in the multiple frames of monitoring images is greater than the average number of times other non-aggregated persons appear in the multiple frames of monitoring images when medium-risk human behavior is identified in the multiple frames of monitoring images. If so, a warning signal is generated; when high-risk human behavior is identified in the multiple frames of monitoring images, a warning signal is directly generated, and the warning signal includes: multiple frames of monitoring images, monitoring equipment location information and risk level.

7. A campus safety risk monitoring system according to claim 6, characterized in that: The YOLOv8-CNN-LSTM model in the risk prediction unit is specifically used to: Input multiple frames of surveillance images into the YOLOv8 network in sequence to obtain an object detection frame with annotated object detection frames and corresponding IDs. The object detection frames include: a human body frame and a human skeleton frame; Input the object detection block diagram into the CNN network to extract the feature vector; Input the feature vector into the LSTM network to obtain a time series feature vector; The time series feature vector is classified through a fully connected layer, and the classification includes: low risk, medium risk and high risk.

8. A campus safety risk monitoring system according to claim 7, characterized in that: The hierarchical early warning unit is specifically used to: When a person with medium risk is identified in multiple frames of surveillance footage, the object detection frame diagram of the multiple frames of surveillance footage is obtained, and face recognition is performed on each ID object detection frame in the object detection frame diagram. When a person of key concern is identified, the corresponding ID is marked as the target ID, and the number of occurrences of the target ID in multiple object detection frame diagrams is counted; For non-target IDs, the distance between the non-target ID object detection frames in each object detection frame map is calculated, and the average distance is calculated. The non-target IDs corresponding to the object detection frames in multiple object detection frame maps whose distances are less than the average distance are marked as clustered IDs, and the other non-target IDs are marked as non-clustered IDs; Count the number of occurrences of non-clustered IDs in multiple object detection frame maps, and calculate the average number of occurrences of each non-clustered ID as the medium-risk warning threshold; The number of occurrences of the target ID in the plurality of object detection frame maps is compared with the medium risk warning threshold, and if the number is greater than the medium risk warning threshold, a warning signal is generated.

Images