Campus abnormal behavior detection system based on artificial intelligence

Abstract

The invention discloses a campus abnormal behavior detection system based on artificial intelligence, and relates to the technical field of abnormal behavior detection, and the system comprises the steps: connecting a campus monitoring platform, and determining a first monitoring blind area and a first monitoring neighborhood; establishing a first standard aggregation feature set of the first monitoring blind area; obtaining first monitoring video data; identifying and marking personnel, and generating a first monitoring object time sequence; performing aggregation feature recognition to generate a first real-time aggregation feature; performing abnormal aggregation feature recognition to generate a first abnormal probability; and when the first abnormal probability is greater than a preset early warning probability, generating first abnormal early warning information. The technical problems that in the prior art, traditional campus monitoring is difficult to cover a blind area, and abnormal gathering behaviors in different time periods cannot be accurately judged are solved, and the technical effects that the abnormal behaviors in the campus monitoring blind area are effectively detected and pre-warned, and the campus safety management level is improved are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal behavior detection, and in particular to a campus abnormal behavior detection system based on artificial intelligence. Background Art

[0002] In today's society, campus safety has always been a focus of great concern to all parties. With the continuous development of education and the gradual expansion of campus scale, the campus environment is becoming increasingly complex. Although the traditional campus monitoring system has guaranteed the safety of the campus to a certain extent, it has many limitations. On the one hand, there are inevitably some monitoring blind spots on campus. These areas are difficult to be directly covered by conventional monitoring equipment due to their special location or equipment layout. In these monitoring blind spots, once abnormal behaviors occur, such as student conflicts and illegal intrusions, traditional monitoring systems often cannot detect them in time, which brings great hidden dangers to campus safety. On the other hand, traditional monitoring systems lack effective analysis methods for the activities of people in different time zones on campus. The gathering of people on campus will vary greatly in different time periods. For example, people are more concentrated during breaks, while they are relatively dispersed during classes. If these gathering characteristics of people cannot be accurately analyzed, it is difficult to determine whether there is abnormal gathering behavior and it is impossible to take corresponding measures to intervene in time.

[0003] The existing technology has the technical problem that traditional campus monitoring is difficult to cover blind spots and cannot accurately judge abnormal gathering behaviors at different time periods. Summary of the invention

[0004] The present application provides an artificial intelligence-based campus abnormal behavior detection system, which is used to solve the technical problems in the prior art that traditional campus monitoring is difficult to cover blind spots and cannot accurately judge abnormal aggregation behaviors in different time periods.

[0005] In view of the above problems, this application provides an artificial intelligence-based campus abnormal behavior detection system.

[0006] The present application provides an artificial intelligence-based campus abnormal behavior detection system, the system comprising:

[0007] A first monitoring blind spot determination module, the first monitoring blind spot determination module is used to connect to the campus monitoring platform, determine the first monitoring blind spot, and the first monitoring neighborhood connected to the first monitoring blind spot; a first standard aggregation feature set establishment module, the first standard aggregation feature set establishment module is used to establish a first standard aggregation feature set for the first monitoring blind spot, wherein the first standard aggregation feature set includes aggregation features of multiple time zones; a first monitoring video data acquisition module, the first monitoring video data acquisition module is used to acquire the first monitoring video data monitored by the monitoring equipment in the first monitoring neighborhood; a first monitoring object timing generation module, the first monitoring object timing generation module is based on the first The monitoring video data identifies and marks the personnel entering the first monitoring blind spot from the first monitoring neighborhood, and generates a first monitoring object time series; a first real-time aggregation feature generation module, the first real-time aggregation feature generation module identifies the aggregation feature based on the first monitoring object time series, and generates a first real-time aggregation feature; a first abnormal probability generation module, the first abnormal probability generation module is used to compare the first real-time aggregation feature with the first standard aggregation feature set, identify the abnormal aggregation feature, and generate a first abnormal probability; a first abnormal warning information generation module, the first abnormal warning information generation module is used to generate a first abnormal warning information when the first abnormal probability is greater than a preset warning probability.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] The first monitoring blind area determination module is used to connect to the campus monitoring platform, determine the first monitoring blind area, and the first monitoring neighborhood connected to the first monitoring blind area; the first standard aggregation feature set establishment module is used to establish the first standard aggregation feature set of the first monitoring blind area; the first monitoring video data acquisition module is used to acquire the first monitoring video data monitored by the monitoring equipment in the first monitoring neighborhood; the first monitoring object time sequence generation module identifies and marks the personnel and generates the first monitoring object time sequence; the first real-time aggregation feature generation module performs aggregation feature recognition based on the first monitoring object time sequence and generates the first real-time aggregation feature; the first abnormal probability generation module is used to compare the first real-time aggregation feature with the first standard aggregation feature set, perform abnormal aggregation feature recognition, and generate the first abnormal probability; the first abnormal warning information generation module is used to generate the first abnormal warning information when the first abnormal probability is greater than the preset warning probability. The technical effect of realizing effective detection and warning of abnormal behaviors in the campus monitoring blind area and improving the level of campus safety management is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A structural diagram of an artificial intelligence-based campus abnormal behavior detection system is provided for the embodiment of the present application;

[0012] Figure 2 A schematic diagram of a process for determining a first monitoring blind spot in an artificial intelligence-based campus abnormal behavior detection system is provided for an embodiment of the present application.

[0013] Explanation of the accompanying drawings: first monitoring blind spot determination module 10, first standard aggregation feature set establishment module 20, first monitoring video data acquisition module 30, first monitoring object time series generation module 40, first real-time aggregation feature generation module 50, first abnormal probability generation module 60, first abnormal warning information generation module 70. DETAILED DESCRIPTION

[0014] The present application provides an artificial intelligence-based campus abnormal behavior detection system to solve the technical problems in the prior art that traditional campus monitoring is difficult to cover blind spots and cannot accurately judge abnormal gathering behaviors at different time periods.

[0015] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0016] Examples, such as Figure 1 As shown, the present application provides an artificial intelligence-based campus abnormal behavior detection system, the system comprising:

[0017] The first monitoring blind area determination module 10 is used to connect to the campus monitoring platform, determine the first monitoring blind area, and the first monitoring neighborhood connected to the first monitoring blind area.

[0018] Specifically, the first monitoring blind spot determination module 10 is responsible for connecting to the campus monitoring platform. By establishing a connection with the campus monitoring platform, the module can obtain the layout information, monitoring range, real-time monitoring data and other related system parameters of each monitoring device in the campus. This connection is achieved through network communication technology, ensuring that the module can receive information from the monitoring platform in a timely and accurate manner. Connecting the campus monitoring platform provides a data basis for determining the first monitoring blind spot and the first monitoring neighborhood. Determining the first monitoring blind spot is one of the core tasks of the module. There may be some areas in the campus that cannot be directly covered by the monitoring equipment due to various reasons. These areas are defined as monitoring blind spots. The module will comprehensively consider factors such as the installation location, monitoring angle, coverage, and building layout of the monitoring equipment, and determine the specific location and range of the first monitoring blind spot through algorithm analysis or manual judgment. For example, analyze the field of view of the monitoring equipment to find out those areas that are not covered by any monitoring equipment; or consider the occlusion of the building to determine the location where the blind spot may exist. Determining the first monitoring blind spot is crucial for campus safety management, because these areas may become hidden points for safety hazards and require special attention. At the same time, the module also needs to determine the first monitoring neighborhood connected to the first monitoring blind spot, take the first monitoring blind spot as the center, analyze which surrounding monitoring areas are adjacent to the blind spot, and determine the neighborhood by calculating the distance relationship between the coverage of the monitoring device and the blind spot. For example, if the coverage of a monitoring device overlaps with the boundary of the blind spot to a certain extent, then the area corresponding to this monitoring device can be considered as part of the first monitoring neighborhood. The purpose of determining the first monitoring neighborhood is to use the monitoring data of the neighborhood to infer the situation of the blind spot. Since the blind spot cannot be directly monitored, by analyzing the personnel activities, environmental changes and other information in the neighborhood, it is possible to indirectly understand the situation that may occur in the blind spot, thereby improving the accuracy and comprehensiveness of abnormal behavior detection.

[0019] The first standard aggregation feature set establishing module 20 is used to establish a first standard aggregation feature set for the first monitoring blind spot, wherein the first standard aggregation feature set includes aggregation features of multiple time zones.

[0020] Specifically, the first standard aggregation feature set establishment module 20 is dedicated to constructing the first standard aggregation feature set for the first monitoring blind area. This feature set covers the aggregation features of multiple time zones, including the number of people gathered under normal circumstances and the personnel change characteristics that change over time. In order to obtain these features, the module will collect the regular behavior samples of the first monitoring blind area in multiple time zones. These samples come from the historical data of the campus monitoring platform, including monitoring videos and personnel activity records of different time periods. By analyzing these samples, the normal behavior patterns of the first monitoring blind area at different times are understood. Then, based on these regular behavior samples, feature extraction is performed. For the determination of the threshold of the number of gathered people, the module will count the range of the number of people appearing in the samples of different time zones. For example, in the time zone before class in the morning, a certain number of students will pass through the first monitoring blind area to the classroom. By analyzing the samples before class in the morning, the threshold of the number of gathered people in this time zone can be determined. This threshold represents the maximum number of people that may appear in the first monitoring blind area of ​​this time zone under normal circumstances. For the stable feature of the change rate of gathered people, it represents the stability of the change rate of the number of gathered people. The module will calculate the rate of change of the number of people in samples of different time zones over time. If the number of people in a certain time zone changes relatively slowly and the rate of change fluctuates slightly, then it can be considered that the stability characteristic of the change rate of the aggregated people in this time zone is high. On the contrary, if the number of people changes dramatically and the rate of change fluctuates greatly, then the stability characteristic is low. By analyzing multiple samples, the stability characteristic of the change rate of the aggregated people in each time zone can be determined.

[0021] The first monitoring video data acquisition module 30 is used to acquire the first monitoring video data monitored by the monitoring equipment in the first monitoring neighborhood.

[0022] Specifically, the main function of the first monitoring video data acquisition module 30 is to obtain the first monitoring video data monitored by the monitoring equipment in the first monitoring neighborhood. Since the first monitoring blind area cannot be monitored directly, by obtaining the monitoring video data of its neighborhood, an important source of information can be provided for inferring the situation of the blind area. First, the module will establish a connection with the monitoring equipment in the first monitoring neighborhood, which is realized through a network communication protocol or a specific monitoring system interface. Once the connection is successfully established, the module can receive video data from these monitoring devices in real time. Then, the module will perform preliminary processing and storage on the acquired video data, and perform operations such as compression and encoding on the video to reduce the amount of data and facilitate subsequent analysis and processing. At the same time, the video data is stored in a specific database or storage device for easy call and analysis at any time. Obtaining the monitoring video data of the first monitoring neighborhood, on the one hand, the video of the neighborhood provides information about the flow of people. If there are people entering the first monitoring blind area from the neighborhood, the movement trajectory of these people can be tracked by analyzing the video of the neighborhood, thereby inferring their possible behavior in the blind area. On the other hand, the video of the neighborhood can also reflect the changes in the environment. For example, if there is an abnormal crowd gathering, noise or other abnormal phenomena in the neighborhood, it means that there are potential problems in the first monitoring blind spot. The first monitoring video data acquisition module 30 obtains the first monitoring video data monitored by the monitoring equipment in the first monitoring neighborhood, which provides important basic data for the campus abnormal behavior detection system, and helps to improve the accuracy and timeliness of abnormal behavior detection in the first monitoring blind spot.

[0023] The first monitoring object timing generation module 40 identifies and marks the persons who enter the first monitoring blind spot from the first monitoring neighborhood based on the first monitoring video data, and generates a first monitoring object timing.

[0024] Specifically, the first monitoring object time sequence generation module 40 is based on the first monitoring video data, focusing on identifying and marking the personnel entering the first monitoring blind spot from the first monitoring neighborhood, and then generating the first monitoring object time sequence. First, the first monitoring video data is received, which comes from the monitoring equipment in the first monitoring neighborhood and records the dynamic situation of the neighborhood and the surrounding area. Based on the neural network model of deep learning, the position and outline of the personnel in the video screen are accurately detected. Once the personnel are identified, it is further determined whether these personnel enter the first monitoring blind spot from the first monitoring neighborhood, and a comprehensive judgment is made in combination with the information such as the movement direction, position change and layout of the monitoring equipment in the video. For the personnel determined to enter the first monitoring blind spot from the first monitoring neighborhood, the annotation is performed, and the annotation includes the identity information of the personnel (if identifiable), the entry time, the entry position, etc. Through the annotation, the specific situation of each person entering the blind spot can be clearly recorded. Finally, according to the annotated information, the first monitoring object time sequence is generated. This time sequence is a record arranged in chronological order, which shows in detail the situation of the personnel entering the first monitoring blind spot from the first monitoring neighborhood at different time points. For example, the time series can record that person A enters the blind spot from a certain location in the neighborhood at time point T1, and person B enters the blind spot at time point T2, etc. By analyzing this time series, we can understand the pattern and frequency of people entering the blind spot, as well as the relationship between people. If the number of people entering the blind spot increases abnormally or a specific combination of people appears in a certain period of time, it means that there is a potential abnormal situation. At the same time, the time series can also help trace the occurrence process of abnormal behavior and provide strong evidence for subsequent investigation and processing.

[0025] The first real-time aggregation feature generation module 50 performs aggregation feature recognition based on the first monitored object time series to generate a first real-time aggregation feature.

[0026] Specifically, first, the generated first monitoring object time series is received. This time series records the situation of people entering the first monitoring blind spot from the first monitoring neighborhood at different time points, providing basic data for the identification of aggregation features. Then, the aggregation feature identification is started. The aggregation feature mainly reflects the aggregation of people in the first monitoring blind spot at a specific time. For the generation of the first real-time aggregation feature, multiple aspects will be considered. On the one hand, the number of people in the first monitoring blind spot at the current moment is counted. By analyzing the first monitoring object time series, it is determined how many people are in the blind spot at a specific time point. This is an important aggregation feature indicator. On the other hand, attention is paid to the distribution and dynamic changes of people. In addition, combined with the time factor, the residence time of people in the blind spot and the change trend of the number of people over time are analyzed. If a large number of people rush into the blind spot in a short time and stay for a long time, it means that an abnormal situation has occurred. Finally, through the comprehensive analysis of these factors, the first real-time aggregation feature is generated. This feature is represented by a feature vector or a set of descriptive indicators, such as the number of people at the current moment, the distribution density of people, the change rate of people, etc. These real-time aggregation features will be compared with the first standard aggregation feature set to determine whether there is any abnormality in the first monitoring blind area.

[0027] The first abnormal probability generating module 60 is used to compare the first real-time aggregation feature with the first standard aggregation feature set, perform abnormal aggregation feature recognition, and generate a first abnormal probability.

[0028] Specifically, the first real-time aggregation feature is obtained, which is an immediate reflection of the current gathering of people in the first monitoring blind spot, including the current number of people, distribution status, and change trend. At the same time, the first standard aggregation feature set that has been established is also obtained, which covers the normal aggregation features of the first monitoring blind spot under multiple time zones, providing a standard reference for judging whether the current situation is abnormal. Next, the first real-time aggregation feature is compared with the features of the corresponding time zone in the first standard aggregation feature set. The current actual number of people is compared with the threshold of the normal number of people gathered in the time zone in the standard set. If it exceeds the threshold range, there is an abnormality; the personnel change rate is also compared with the stable characteristics of the personnel change rate under normal circumstances. If the change rate fluctuates greatly or deviates from the normal range, there is an abnormality. Through this comparison, the abnormal aggregation feature is identified. If there is a large difference between the real-time feature and the standard feature, it means that an abnormal situation has occurred in the current first monitoring blind spot. These abnormal features are manifested as an abnormal increase or decrease in the number of people, abnormal concentration or dispersion of personnel distribution, and abnormal fluctuations in the personnel change rate. Finally, according to the recognition result of the abnormal aggregation feature, the first abnormal probability is generated. This probability value reflects the possibility of an abnormal situation in the first monitoring blind spot. When the probability of abnormality is high, early warning information will be issued in a timely manner to remind relevant personnel to pay close attention to the first monitoring blind spot so that appropriate measures can be taken to ensure campus safety.

[0029] The first abnormal warning information generating module 70 is used to generate first abnormal warning information when the first abnormal probability is greater than a preset warning probability.

[0030] Specifically, if the first abnormal probability exceeds the preset warning probability, it indicates that a more serious abnormal situation may occur in the first monitoring blind spot. At this time, the module will start quickly and generate the first abnormal warning information. This warning information is presented in various forms, such as popping up a warning window on the display screen of the campus monitoring center, sending text messages or emails to notify relevant management personnel, or triggering an alarm. The content of the warning information includes the location where the abnormality occurs (i.e., the first monitoring blind spot), the type of abnormality (judged according to the specific circumstances of the abnormal probability, such as too many people gathering, abnormal personnel change rate, etc.) and the recommended measures. In this way, after receiving the warning information, the relevant personnel can quickly understand the specific circumstances of the abnormal situation and take corresponding countermeasures in time, such as sending security personnel to the scene to check and start the emergency plan, which helps to timely discover and deal with abnormal situations on campus and ensure the safety of teachers and students and the normal order of the campus.

[0031] In one possible implementation, Figure 2 As shown, the first monitoring blind area determination module 10 also includes:

[0032] Taking the first monitoring blind area as the center, identifying the boundary monitoring device located in the first monitoring blind area; acquiring the monitoring parameters of the boundary monitoring device; determining the boundary monitoring range and monitoring direction based on the monitoring parameters, and generating the first monitoring neighborhood.

[0033] Specifically, in the campus abnormal behavior detection system, with the first monitoring blind spot as the center, boundary detection is first required. This process is based on the campus monitoring platform. By analyzing the field of view of the existing monitoring cameras, the exact boundary of the blind spot is determined. Next, the monitoring equipment located at the edge of the blind spot is identified, including cameras and sensors, and their positions and viewing angles are recorded. By integrating the information of these devices and comparing the field of view of the boundary monitoring equipment with the first monitoring blind spot, the monitoring area that can cover the blind spot can be identified. Finally, this identification result will be fed back to the monitoring system to optimize the monitoring layout and ensure effective coverage of the monitoring blind spot, thereby improving the campus security prevention capabilities.

[0034] After identifying the boundary monitoring devices, the monitoring parameters of these devices are obtained, such as the camera's viewing angle, focal length, resolution, and height. These parameters will be used to determine whether the monitored object enters or leaves the first monitoring blind spot, analyze the viewing angle range of the device, and determine its relative position in combination with the object's movement trajectory in the video. If the object's movement direction matches the monitoring range of the boundary device, its behavior can be determined as entering or leaving the blind spot. In this way, the system can monitor the campus security status in real time and issue early warnings in a timely manner.

[0035] Based on the monitoring parameters of the acquired boundary monitoring devices, the viewing angle and monitoring range of each device are analyzed to determine its coverage area. By calculating the monitoring direction of each device, the overlapping monitoring areas are identified to form a complete first monitoring neighborhood. On this basis, the boundaries of the neighborhood are further optimized in combination with the height and focal length of the device to ensure that there are no blind spots. In addition, the generated first monitoring neighborhood will be marked as a monitorable area, which can effectively track objects entering or leaving the blind spot, thereby improving the accuracy and responsiveness of the overall security monitoring.

[0036] In a possible implementation, the first standard aggregation feature set establishing module 20 further includes:

[0037] Obtain regular behavior samples of the first monitoring blind area in the multiple time zones; perform feature extraction based on the regular behavior samples to generate clustering features of the multiple time zones, wherein the clustering features include a threshold value for the number of gathered people and a stable feature of the rate of change of gathered people; the stable feature of the rate of change of gathered people indicates the stability of the rate of change of the number of gathered people.

[0038] Specifically, in order to obtain samples of regular behavior in multiple time zones of the first monitoring blind spot, data mining is performed by analyzing historical monitoring video data. First, the monitoring videos in different time periods are extracted to identify and classify common activity types in the blind spot, such as people passing, gathering or wandering. Then, the frequency, duration and occurrence pattern of the behavior in each time zone are recorded to establish a regular behavior database. These samples provide reference standards for subsequent anomaly detection, so as to effectively identify behavioral patterns that deviate from the norm, thereby improving the ability to respond to abnormal behaviors.

[0039] Based on the obtained routine behavior samples, feature extraction is performed to generate clustering features for multiple time zones. First, the behavior data in different time periods are analyzed to extract the number of people gathering and their changing trends. For each time zone, the threshold of the number of people gathering is calculated, that is, the minimum number of people considered to be gathering in that time zone. This threshold is set according to the average level and standard deviation of historical data to ensure that routine behavior can be accurately reflected. In addition, the rate of change of the gathering of people is analyzed to extract stable features. This includes calculating the rate of increase and decrease of the number of people gathering in a specific time period to identify normal fluctuations and abnormal surges or decreases. Through these clustering features, a dynamic monitoring model is established to effectively identify abnormal gathering behaviors in real-time monitoring and provide timely warnings for campus safety.

[0040] The stable feature of the change rate of the number of people gathered is used to measure the stability and consistency of the change in the number of people gathered. This feature reflects the regularity of gathering behavior by analyzing the rate of change of the number of people gathered in a specific time period. The number of people gathered at different time points (such as every minute, every hour) is recorded, and the rate of change, that is, the increase or decrease in the number of people per unit time, is calculated. In order to quantify stability, a statistical method is used to calculate the standard deviation of the rate of change. If the standard deviation of the rate of change is small, it means that the fluctuation of the number of people gathered in this time zone is small and the change is relatively stable; on the contrary, if the standard deviation is large, it means that the change in the number of people gathered is more drastic and there is a risk of abnormal behavior. In addition, a threshold value for the rate of change is established. The rate of change exceeding this threshold will be considered unstable, which indicates abnormal gathering or personnel flow. By monitoring this stable feature, potential safety hazards can be better identified, providing strong data support for campus management.

[0041] In a possible implementation, the first abnormal probability generating module 60 further includes:

[0042] Obtain the first monitoring time zone of the first monitoring video data; match the target standard aggregation feature in the first standard aggregation feature set according to the first monitoring time zone; compare the first real-time aggregation feature with the matching target standard aggregation feature, establish the aggregation feature deviation, and fuse the abnormal weights in combination with the monitoring object to generate the first abnormal probability.

[0043] Specifically, first extract the relevant surveillance video data from the monitoring equipment. This process ensures that the acquired data is complete and has not been tampered with. Analyze the timestamp information in the video data to determine the specific recording time period, which is achieved by extracting the metadata of the video file. According to the set time range, the timestamps are classified to divide the specific first monitoring time zone, such as dividing a day into different time periods such as morning, afternoon and evening. Finally, confirm the identified monitoring time zone to prepare for subsequent cluster feature matching and abnormal behavior detection. This process ensures the accurate grasp of the monitoring time zone and helps to improve the effectiveness of subsequent data analysis.

[0044] According to the first monitoring time zone, the first standard aggregation feature set will be analyzed to match the target standard aggregation feature. First, the real-time aggregation features in the time zone are extracted, including the number of people currently gathered, the type of activity, and the trend of change. Then, these real-time features are compared with the historical behavior patterns in the first standard aggregation feature set to find the most similar features. This matching process involves calculating the similarity between features and using the Euclidean distance algorithm to determine the degree of difference between the current aggregation features and historical routine behaviors. Through this matching, the target standard aggregation features are identified as a reference benchmark for subsequent anomaly detection. This process not only helps to clarify the aggregation behavior of the current monitoring time zone, but also provides key data support for identifying potential anomalies, thereby improving the intelligence and responsiveness of the monitoring system.

[0045] Determine the first real-time aggregation feature and the target standard aggregation feature that matches the current time zone. For the first monitoring blind spot, clarify the real-time aggregation features such as the actual number of people and the rate of change of people at the current moment. At the same time, find the standard aggregation personnel number threshold and the standard personnel change rate stability feature of the corresponding time zone from the first standard aggregation feature set. Next, calculate the aggregation feature deviation, and calculate the personnel number deviation, that is, the absolute value of the difference between the actual number of people and the standard number of people threshold; and the personnel change rate deviation, which is the absolute value of the difference between the actual number of people and the standard number of people change rate stability feature. Then, the two deviations are combined by the Euclidean distance to obtain the aggregation feature deviation value. For each monitoring object entering the first monitoring blind spot, determine the abnormal weight according to its historical campus abnormal behavior record, and count the number of abnormal behaviors of each monitoring object in history The ratio of the total number of behavior records is used as the abnormal weight of the monitoring object. After that, calculate the sum of the abnormal weights of all monitoring objects, and then divide the abnormal weight of each monitoring object by this sum for normalization. When calculating the first abnormality probability, the importance coefficients of the clustered feature deviation and the abnormality weight are set, and the sum of the two coefficients is 1. For each monitored object, an adjustment factor is set, which is determined based on factors such as the severity of historical abnormal behavior. The first abnormality probability is equal to the clustered feature deviation multiplied by its importance coefficient, plus the sum of the products of the normalized abnormal weights and the adjustment factor of all monitored objects multiplied by the importance coefficient of the abnormal weight. In this way, the weighted average algorithm is used to accurately calculate the probability of abnormal situations occurring in the first monitoring blind spot, providing strong support for campus safety management.

[0046] In a possible implementation, the first abnormal probability generating module 60 further includes:

[0047] Initialize the abnormal weights of all objects entering the first monitoring blind spot in the first monitoring object time series to generate an initialized abnormal weight set; identify the initial abnormal probability based on the clustered feature deviation, wherein the initial abnormal probability is proportional to the size of the clustered feature deviation; perform weighted influence calculation on the initial abnormal probability using the initialized abnormal weight set to generate the first abnormal probability.

[0048] Specifically, the abnormal weights of all objects that enter the first monitoring blind spot in the first monitoring object time series are initialized. At this stage, an initial abnormal weight is assigned to each object that enters the blind spot. This weight is determined based on some basic factors, such as the type of object (student, teacher, staff, etc.), the frequency of entering the blind spot, etc. By initializing the abnormal weights of all objects, an initialized abnormal weight set is generated.

[0049] Next, the initial abnormal probability is identified based on the aggregation feature deviation. The aggregation feature deviation is established by comparing the first real-time aggregation feature with the matching target standard aggregation feature, which reflects the difference between the current state and the normal state of the first monitoring blind spot. The initial abnormal probability is proportional to the size of the aggregation feature deviation, that is, the larger the deviation, the higher the initial abnormal probability, because a larger deviation means that the current blind spot situation is much different from the normal situation, and the possibility of abnormality increases accordingly.

[0050] Finally, the initial abnormal probability is weighted and calculated with the initialized abnormal weight set to generate the first abnormal probability. In this step, the abnormal weight of each object will affect the initial abnormal probability. If the abnormal weight of an object is high, its contribution to the first abnormal probability will be relatively large. Through this weighted calculation, the degree of influence of different objects on the abnormal situation in the blind spot can be more accurately reflected. For example, if there are some objects that are often associated with abnormal situations in history, their high abnormal weights will increase the first abnormal probability accordingly, thereby attracting more attention from the system. By initializing the abnormal weights of the objects entering the first monitoring blind spot, identifying the initial abnormal probability based on the clustering feature deviation, and then performing a weighted influence calculation, a more accurate and targeted first abnormal probability is generated, providing an important basis for abnormal behavior detection and early warning on campus.

[0051] In a possible implementation, the first abnormal probability generating module 60 further includes:

[0052] Obtain a historical campus abnormal behavior library, wherein any abnormal behavior sample in the historical campus abnormal behavior library includes an abnormal person identification sample; based on the abnormal person identification sample, count the first abnormal behavior frequency characteristics of any first person who has exhibited abnormal behavior; based on the first abnormal behavior frequency characteristic configuration, initialize the abnormal weight of the first person to generate a first abnormal weight; when the object entering the first monitoring blind spot includes the first person, add the first abnormal weight to the initialized abnormal weight set.

[0053] Specifically, obtaining the historical campus abnormal behavior database is an important step. This historical campus abnormal behavior database stores information about various abnormal behaviors that have occurred in the past. Among them, any abnormal behavior sample contains abnormal person identification samples. These identification samples clearly indicate the specific people involved in the abnormal behavior. By obtaining this historical campus abnormal behavior database, past abnormal behaviors are analyzed and studied, so as to better identify and predict abnormal situations that may occur in the future. For example, by analyzing abnormal person identification samples, we can understand which people have participated in abnormal behaviors, as well as the specific characteristics and patterns of these abnormal behaviors, which provides an important reference for subsequent abnormal behavior detection and early warning.

[0054] The frequency of abnormal behavior of the first person is counted in detail, including the number of times the first person has abnormal behavior, the frequency, and the distribution of these behaviors in different time periods. By analyzing these data, the abnormal behavior pattern of the first person can be understood, such as whether there is a specific time pattern, whether it is related to certain specific events or situations, etc. This statistical analysis can provide an important basis for subsequent abnormal behavior detection. According to the frequency characteristics of the first abnormal behavior, the possibility of the first person having abnormal behavior again can be judged, and the degree of attention and warning level can be adjusted accordingly. At the same time, these characteristics also help the system discover potential abnormal behavior trends and take measures to prevent and intervene in advance, thereby improving the safety and stability of the campus.

[0055] The process of initializing the abnormal weight is to give the first person a value that can reflect his abnormal behavior tendency. This value is the first abnormal weight. If the first person's abnormal behavior frequency is high, then the corresponding first abnormal weight will increase accordingly, indicating that the person is more likely to have abnormal behavior and needs more attention. On the contrary, if the first person's abnormal behavior frequency is low, then the first abnormal weight will be relatively small. The first abnormal weight generated in this way will play an important role in the subsequent abnormal probability calculation and warning generation. It will comprehensively consider the first abnormal weight of each person and other relevant factors to more accurately judge whether there is an abnormal situation on campus, and issue early warning information in time to ensure the safety and order of the campus.

[0056] When an object enters the first monitoring blind spot, real-time monitoring and analysis are performed. If the object entering the blind spot includes an individual previously marked as the first person, the first abnormal weight corresponding to the first person will be added to the initialized abnormal weight set. By adding the first abnormal weight to the set, the historical abnormal behavior of the person entering the blind spot is more comprehensively considered. The first abnormal weight reflects the possibility or tendency of the first person to exhibit abnormal behavior. Adding it to the set can more accurately assess the abnormal degree of the current situation in the subsequent abnormal probability calculation. For example, if the first person has exhibited abnormal behavior many times in history and his first abnormal weight is high, then when he enters the first monitoring blind spot, he will pay more attention to this situation because his presence may increase the risk of abnormal situations in the blind spot. In this way, potential abnormal situations can be more effectively identified and warned, and the safety and management efficiency of the campus can be improved.

[0057] In a possible implementation, the first abnormal probability generating module 60 further includes:

[0058] If the object entering the first monitoring blind spot does not appear in the historical campus abnormal behavior library, the corresponding abnormal weight is initialized to zero.

[0059] Specifically, for objects entering the first monitoring blind spot, a check will be conducted to determine whether they have appeared in the historical campus abnormal behavior library. If an object does not appear in the library, it means that there is no record of the object's previous abnormal behavior. In this case, in order to accurately assess the abnormal situation, the abnormal weight of the object that does not appear in the historical library is initialized. Because there is no historical data to support the possibility of abnormal behavior of the object, giving it a zero weight can avoid misjudgment of the abnormal probability. In this way, when calculating the first abnormal probability, it will not be disturbed by these objects without historical abnormal records, and can more accurately focus on those objects with potential abnormal risks, thereby improving the accuracy and reliability of abnormal behavior detection.

[0060] In a possible implementation, the first abnormal probability generating module 60 further includes:

[0061] Obtain the non-zero weights in the initialized abnormal weight set; calculate the proportion coefficient of the non-zero weights; when the proportion coefficient is greater than a preset proportion coefficient threshold, weight the initial abnormal probability after superimposing the non-zero weights; when the proportion coefficient is less than or equal to the preset proportion coefficient threshold, weight the initial abnormal probability with a maximum weight value to generate the first abnormal probability.

[0062] Specifically, the initialized abnormal weight set is traversed to filter out the elements whose weight values ​​are not zero. These non-zero weights represent that the corresponding monitoring objects have abnormal behavior records in the historical campus abnormal behavior library. By obtaining these non-zero weights, the possibility of abnormal behavior in the current monitoring scene can be more accurately evaluated, providing key information support for subsequent abnormal probability calculation and early warning decision-making.

[0063] Calculate the proportion coefficient of non-zero weights and compare it with the preset proportion coefficient threshold. If the proportion coefficient is greater than the preset proportion coefficient threshold, it means that the non-zero weight accounts for a large proportion of the whole. At this time, these non-zero weights are superimposed, and the superimposed result will be used to weight the initial abnormal probability. In this way, the influence of multiple objects with abnormal weights on the abnormal probability is fully taken into account, so that the calculated first abnormal probability can more accurately reflect the actual situation. When the proportion coefficient is less than or equal to the preset proportion coefficient threshold, it means that the proportion of non-zero weights in the whole is small. In this case, the initial abnormal probability is weighted with the maximum weight value. The maximum weight value is the maximum weight value that appears in the initial abnormal weight set. The purpose of doing this is to give a certain adjustment to the initial abnormal probability when the non-zero weight ratio is small, so as to ensure the accuracy of the first abnormal probability. By selecting different weighting methods according to the size of the proportion coefficient, the first abnormal probability is reasonably generated, thereby improving the accuracy and reliability of abnormal behavior detection on campus.

[0064] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0066] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. The campus abnormal behavior detection system based on artificial intelligence is characterized by: include: A first monitoring blind area determination module, the first monitoring blind area determination module is used to connect to the campus monitoring platform, determine a first monitoring blind area, and a first monitoring neighborhood connected to the first monitoring blind area; A first standard aggregation feature set establishment module, the first standard aggregation feature set establishment module is used to establish a first standard aggregation feature set of the first monitoring blind area, wherein the first standard aggregation feature set includes aggregation features of multiple time zones; A first monitoring video data acquisition module, the first monitoring video data acquisition module is used to acquire first monitoring video data monitored by the monitoring equipment in the first monitoring neighborhood; A first monitoring object timing generation module, which identifies and marks the persons who enter the first monitoring blind spot from the first monitoring neighborhood based on the first monitoring video data, and generates a first monitoring object timing; A first real-time aggregation feature generation module, which performs aggregation feature recognition based on the first monitored object time series to generate a first real-time aggregation feature; A first abnormal probability generation module, the first abnormal probability generation module is used to compare the first real-time aggregation feature with the first standard aggregation feature set, perform abnormal aggregation feature recognition, and generate a first abnormal probability; The first abnormal warning information generation module is used to generate first abnormal warning information when the first abnormal probability is greater than a preset warning probability.

2. The campus abnormal behavior detection system based on artificial intelligence as claimed in claim 1 is characterized in that: The first monitoring blind area determination module also includes: Taking the first monitoring blind area as the center, identifying the boundary monitoring device located in the first monitoring blind area; Acquiring monitoring parameters of the border monitoring device; A boundary monitoring range and a monitoring direction are determined based on the monitoring parameters to generate the first monitoring neighborhood.

3. The campus abnormal behavior detection system based on artificial intelligence as claimed in claim 1 is characterized in that: The first standard aggregation feature set establishment module also includes: Obtaining regular behavior samples of the first monitoring blind area in the multiple time zones; Extracting features based on the conventional behavior samples to generate clustering features of the multiple time zones, wherein the clustering features include a threshold value of the number of clustered personnel and a stable feature of a change rate of clustered personnel; The stable characteristic of the change rate of the number of people gathered represents the stability of the change rate of the number of people gathered.

4. The campus abnormal behavior detection system based on artificial intelligence as claimed in claim 1 is characterized in that: The first abnormal probability generating module also includes: Obtaining a first monitoring time zone of the first monitoring video data; matching a target standard aggregation feature in the first standard aggregation feature set according to the first monitoring time zone; The first real-time aggregation feature is compared with the matching target standard aggregation feature, an aggregation feature deviation is established, and the abnormal weight is fused in combination with the monitored object to generate the first abnormal probability.

5. The campus abnormal behavior detection system based on artificial intelligence as claimed in claim 4 is characterized in that: The first abnormal probability generating module also includes: Initializing abnormal weights for all objects in the first monitoring object time series that enter the first monitoring blind area to generate an initialized abnormal weight set; Identifying an initial abnormality probability based on the cluster feature deviation, wherein the initial abnormality probability is proportional to the magnitude of the cluster feature deviation; The initial abnormality probability is weightedly calculated using the initial abnormality weight set to generate the first abnormality probability.

6. The campus abnormal behavior detection system based on artificial intelligence as claimed in claim 5 is characterized in that: The first abnormal probability generating module also includes: Acquire a historical campus abnormal behavior library, wherein any abnormal behavior sample in the historical campus abnormal behavior library contains an abnormal person identification sample; Counting the first abnormal behavior frequency characteristics of any first person who has exhibited abnormal behavior based on the abnormal person identification sample; Initializing an abnormal weight for the first person based on the first abnormal behavior frequency feature configuration to generate a first abnormal weight; When the object entering the first monitoring blind spot includes the first person, the first abnormal weight is added to the initialized abnormal weight set.

7. The campus abnormal behavior detection system based on artificial intelligence as claimed in claim 6 is characterized in that: If the object entering the first monitoring blind spot does not appear in the historical campus abnormal behavior library, the corresponding abnormal weight is initialized to zero.

8. The campus abnormal behavior detection system based on artificial intelligence as claimed in claim 7 is characterized in that: The first abnormal probability generating module also includes: Obtaining non-zero weights in the initialized abnormal weight set; A coefficient for the non-zero weight; When the proportion coefficient is greater than the preset proportion coefficient threshold, the initial abnormal probability is weighted after superimposing the non-zero weight; when the proportion coefficient is less than or equal to the preset proportion coefficient threshold, the initial abnormal probability is weighted with the maximum weight value to generate the first abnormal probability.

Images