Detailed Description
The application provides an artificial intelligence-based campus abnormal behavior detection system, which is used for solving the technical problems that a blind area is difficult to cover in traditional campus monitoring and abnormal aggregation behaviors in different periods cannot be accurately judged in the prior art.
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. It will be apparent that the described embodiments are only some, but not all, embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Embodiment, as shown in fig. 1, the present application provides a campus abnormal behavior detection system based on artificial intelligence, the system includes:
The first monitoring blind area determining module 10, the first monitoring blind area determining module 10 is used for connecting a campus monitoring platform, determining a first monitoring blind area, and a first monitoring neighborhood connected with the first monitoring blind area.
Specifically, the first monitoring blind area determining module 10 is responsible for connecting with a campus monitoring platform, and by establishing connection with the campus monitoring platform, the module can obtain layout information, monitoring range, real-time monitoring data and other relevant system parameters of each monitoring device in the campus. The connection is realized through a network communication technology, so that the module can timely and accurately receive information from the monitoring platform, and the campus monitoring platform is connected to provide a data base for determining the first monitoring blind area and the first monitoring neighborhood. The method comprises the steps that a first monitoring blind area is determined to be one of core tasks of a module, a plurality of areas which cannot be directly covered by monitoring equipment due to various reasons possibly exist in a campus, the areas are defined as monitoring blind areas, the module comprehensively considers factors such as the installation position, the monitoring angle, the coverage area and the building layout of the monitoring equipment, and the specific position and the range of the first monitoring blind area are determined through an algorithm analysis or manual judgment mode. For example, the field of view of the monitoring devices is analyzed to find out areas that are not covered by any one of the monitoring devices, or the location where a blind area may exist is determined in consideration of the shielding situation of the building. Determining the first blind spot is critical to campus security management because these areas may be hidden points of security risk, requiring special attention. Meanwhile, the module also needs to determine a first monitoring neighborhood connected with the first monitoring blind area, analyze surrounding monitoring areas which are adjacent to the blind area by taking the first monitoring blind area as the center, and determine the neighborhood by calculating the distance relation between the coverage area of the monitoring equipment and the blind area. For example, if a coverage area of a monitoring device has a certain overlapping portion with a boundary of a dead zone, then the area corresponding to the monitoring device can be considered as a part of a first monitoring neighborhood, and the purpose of determining the first monitoring neighborhood is to infer the situation of the dead zone by using monitoring data of the neighborhood.
The first standard aggregation feature set establishing module 20 is configured to establish a first standard aggregation feature set of the first monitoring blind area, where the first standard aggregation feature set includes aggregation features of a plurality of time zones.
Specifically, the first standard aggregate feature set creation module 20 is operative to create a first standard aggregate feature set for a first monitoring blind zone that encompasses aggregate features for a plurality of time zones, including, in particular, aggregate people under normal conditions and person variation features over time. To obtain these features, the module may collect regular behavior samples of the first blind zone in multiple time zones from historical data of the campus monitoring platform, including monitoring videos, personnel activity records, etc. of different time periods. Through analysis of the samples, the normal behavior mode of the first monitoring blind area at different times is known. Next, feature extraction is performed based on these regular behavior samples, and for the determination of the aggregate person number threshold, the module will count the range of person numbers that occur in the different time zone samples. For example, in a time zone before a lesson in the morning, a certain number of students can go to a classroom through a first monitoring blind area, and by analyzing a plurality of samples before the lesson in the morning, the threshold of the number of people gathered in the time zone can be determined. This threshold represents the maximum number of people that can normally occur in the first zone of the time zone. For the aggregate personnel rate of change stabilization feature, it represents the stability of the aggregate personnel rate of change. The module will calculate the rate of change of the number of people in the samples of different time zones over time. If the change in the number of persons is gradual and the fluctuation in the rate of change is small in a certain time zone, it can be considered that the steady characteristic of the rate of change of the aggregated persons in this time zone is high. Conversely, if the number of people varies drastically, the rate of change fluctuates greatly, and the stability characteristics are low. By analyzing the plurality of samples, an aggregate person rate of change stability characteristic for each time zone can be determined.
The first monitoring video data acquisition module 30 is configured to acquire first monitoring video data monitored by a monitoring device in the first monitoring neighborhood, where the first monitoring video data acquisition module 30 is configured to acquire first monitoring video data monitored by the monitoring device in the first monitoring neighborhood.
Specifically, the primary function of the first surveillance video data acquisition module 30 is to acquire first surveillance video data monitored by surveillance devices within the first surveillance neighborhood. Because the first monitoring blind area cannot be directly monitored, important information sources can be provided for deducing the situation of the blind area by acquiring the monitoring video data of the neighborhood of the first monitoring blind area. Firstly, the module establishes connection with monitoring devices in the first monitoring neighborhood through a network communication protocol or a specific monitoring system interface, and once the connection is established successfully, the module can receive video data from the monitoring devices in real time. Then, the module performs preliminary processing and storage on the acquired video data, and performs operations such as compression and encoding on the video, so as to reduce the data volume and facilitate subsequent analysis and processing. At the same time, the video data is stored in a specific database or storage device for ready recall and analysis. And acquiring monitoring video data of the first monitoring neighborhood, wherein on one hand, the video of the neighborhood provides information about the flow of people, and if people enter the first monitoring blind area from the neighborhood, the action tracks of the people can be tracked by analyzing the video of the neighborhood, so that the possible behaviors of the people in the blind area can be estimated. On the other hand, the video of the neighborhood may also reflect the environment change condition, for example, if abnormal crowd gathering, noise or other abnormal phenomena occur in the neighborhood, which means that the first monitoring blind area also has a potential problem. The first monitoring video data obtaining module 30 provides important basic data for the campus abnormal behavior detection system by obtaining the first monitoring video data monitored by the monitoring equipment in the first monitoring neighborhood, and is beneficial to improving the accuracy and timeliness of the first monitoring blind area abnormal behavior detection.
The first monitoring object timing generation module 40 identifies and marks the person entering the first monitoring blind area from the first monitoring neighborhood based on the first monitoring video data, and generates a first monitoring object timing.
Specifically, the first monitoring target timing generation module 40 focuses on identifying and labeling the person entering the first monitoring blind area from the first monitoring neighborhood based on the first monitoring video data, and further generates the first monitoring target timing. First, first monitoring video data are received, wherein the video data come from monitoring equipment in a first monitoring neighborhood, and dynamic conditions of the neighborhood and surrounding areas are recorded. Based on the neural network model of deep learning, the position and the outline of the person in the video picture are accurately detected. Once the personnel are identified, whether the personnel enter the first monitoring blind area from the first monitoring neighborhood is further judged, and comprehensive judgment is carried out by combining the information of the personnel moving direction, the position change, the layout of the monitoring equipment and the like in the video. Labeling is performed on the person who determines to enter the first monitoring blind area from the first monitoring neighborhood, wherein the labeling comprises identity information (if identifiable) of the person, entering time, entering position and the like. Through the marking, the specific condition of each person entering the blind area can be clearly recorded. And finally, generating a first monitoring object time sequence according to the marked information, wherein the time sequence is a record arranged according to a time sequence, and the situation of the personnel entering the first monitoring blind area from the first monitoring neighborhood at different time points is shown in detail. For example, in the time sequence, it may be recorded that the person a enters the blind area from a certain position of the neighborhood at the time point T1, the person B enters the blind area at the time point T2, and the like. By analyzing the time sequence, the rule, frequency, relationship among people and the like of the people entering the blind area can be known. If the number of people entering the blind area is abnormally increased or a specific combination of people occurs within a certain time period, the potential abnormal situation exists. Meanwhile, the time sequence can also help trace the occurrence process of abnormal behaviors, and powerful evidence is provided for subsequent investigation and processing.
The first real-time aggregation feature generation module 50 performs aggregation feature identification based on the first monitoring object time sequence, and generates a first real-time aggregation feature.
Specifically, first, a generated first monitoring object time sequence is received, the time sequence records the conditions of people entering a first monitoring blind area from a first monitoring neighborhood at different time points, and basic data is provided for identifying the aggregation characteristics. Then, the aggregate feature recognition is started, the aggregate features mainly reflect the personnel aggregate condition of the first monitoring blind area at a specific moment, and multiple aspects are considered for generating the first real-time aggregate features. On the one hand, the number of people in the first monitoring blind area at the current moment is counted, and by analyzing the time sequence of the first monitoring object, the number of people in the blind area at a specific time point is determined, which is an important aggregation characteristic index. On the other hand, attention is paid to the distribution situation and dynamic change of the personnel. In addition, in combination with the time factor, the residence time of the person in the blind area and the trend of the change of the number of persons with time are analyzed, and if a large number of persons rush into the blind area and stay for a long time in a short time, this means that an abnormal situation occurs. Finally, by comprehensively analyzing the factors, a first real-time aggregation feature is generated, and the feature is represented by a feature vector or a set of descriptive indexes, such as the number of people, the distribution density of people, the change rate of people and the like at the current moment. The real-time aggregation features are compared with the first standard aggregation feature set to judge whether the first monitoring blind area has abnormal conditions or not.
The first anomaly probability generation module 60 is configured to compare the first real-time aggregation feature with the first standard aggregation feature set, perform anomaly aggregation feature recognition, and generate a first anomaly probability.
Specifically, a first real-time aggregation characteristic is obtained, which is an instant reflection of the current first monitoring blind area personnel aggregation condition and comprises the current personnel number, distribution state, change trend and other information. Meanwhile, a first standard aggregation characteristic set which is already established is also obtained, wherein the normal aggregation characteristics of the first monitoring blind areas in a plurality of time zones are covered, and a standard reference is provided for judging whether the current situation is abnormal. And comparing the first real-time aggregation characteristic with the characteristic of the corresponding time zone in the first standard aggregation characteristic set. Comparing the current actual personnel number with a normal aggregated personnel number threshold value of the time zone in the standard set, if the current actual personnel number exceeds the threshold value range, the abnormal condition exists, and comparing the personnel change rate with the personnel change rate stable characteristic under the normal condition, if the change rate greatly fluctuates or deviates from the normal range, the abnormal condition exists. By this comparison, abnormal aggregation features are identified. If the real-time features and the standard features have larger differences, the abnormal conditions of the first monitoring blind area are indicated, and the abnormal features are represented by abnormal increase or decrease of the number of people, abnormal concentration or dispersion of the distribution of the people, abnormal fluctuation of the change rate of the people and the like. And finally, generating a first abnormal probability according to the identification result of the abnormal aggregation characteristic. This probability value reflects the magnitude of the likelihood of the occurrence of an abnormal condition in the first blind monitoring zone. When the abnormal probability is higher, early warning information is sent timely, and related personnel are reminded to pay close attention to the first monitoring blind area so as to take corresponding measures to ensure campus safety.
The first abnormality early warning information generation module 70 is configured to generate first abnormality early warning information when the first abnormality probability is greater than a preset early warning probability.
Specifically, if the first anomaly probability exceeds the preset early warning probability, the first monitoring blind area is indicated to be possibly in serious anomaly. At this time, the module is started rapidly to generate first abnormality early warning information. The early warning information is presented in various forms, such as popping up a warning window on a display screen of a campus monitoring center, sending a short message or mail to inform related management personnel, triggering an alarm sound and the like. The content of the early warning information includes the position (i.e., the first monitoring blind area) where the abnormality occurs, the type of the abnormality (judged according to the specific situation of the abnormality probability, such as excessive people gathering, abnormal person change rate, etc.), and the recommended measures to be taken. Therefore, after receiving the early warning information, related personnel can quickly know the specific conditions of the abnormal conditions and timely take corresponding countermeasures, such as sending security personnel to the site to check and start an emergency plan, thereby being beneficial to timely finding and processing the abnormal conditions in the campus and ensuring the safety of teachers and students and the normal order of the campus.
In one possible implementation, as shown in fig. 2, the first monitoring blind area determining module 10 further includes:
The method comprises the steps of taking a first monitoring blind area as a center, identifying boundary monitoring equipment positioned in the first monitoring blind area, obtaining monitoring parameters of the boundary monitoring equipment, determining a boundary monitoring range and a monitoring direction based on the monitoring parameters, and generating the first monitoring neighborhood.
Specifically, in the campus abnormal behavior detection system, a first monitoring blind area is taken as a center, boundary detection is needed first, and the accurate boundary of the blind area is determined by analyzing the visual field range of the existing monitoring camera based on a campus monitoring platform. Next, monitoring devices located at the edges of the blind area, including cameras and sensors, etc., are identified, and their positions and viewing angles are recorded. By integrating the information of the devices, the visual field of the boundary monitoring device is compared with the first monitoring blind area, and therefore the monitoring area capable of covering the blind area is identified. And finally, feeding back the identification result to a monitoring system, optimizing the monitoring layout, and ensuring effective coverage of the monitoring blind area so as to improve the campus security and protection capability.
After identifying the boundary monitoring devices, the monitoring parameters of the devices, such as the angle of view, focal length, resolution, height, and the like of the camera are acquired. These parameters will be used to determine whether the monitored object is entering or exiting the first blind zone, analyze the viewing angle range of the device, and determine its relative position in combination with the movement track of the object in the video. If the moving direction of the object coincides with the monitoring range of the boundary device, it can be determined that the object is behaving as entering or leaving the dead zone. By the mode, the system can monitor the campus safety state in real time and timely send out early warning.
Based on the acquired monitoring parameters of the boundary monitoring devices, the viewing angle and the monitoring range of each device are analyzed to determine the coverage area of each device. And identifying the mutually overlapped monitoring areas by calculating the monitoring directions of the devices to form a complete first monitoring neighborhood. On the basis, the boundary of the neighborhood is further optimized by combining the height and the focal length of the equipment so as to ensure no blind area. In addition, the generated first monitoring neighborhood is marked as a monitorable area, and objects entering or leaving the blind area can be effectively tracked, so that the accuracy and the response capability of the overall safety monitoring are improved.
In one possible implementation manner, the first standard aggregate feature set establishing module 20 further includes:
The method comprises the steps of acquiring conventional behavior samples of a first monitoring blind area in a plurality of time zones, extracting features based on the conventional behavior samples, and generating aggregation features of the plurality of time zones, wherein the aggregation features comprise an aggregation personnel number threshold and an aggregation personnel change rate stable feature, and the aggregation personnel change rate stable feature represents the stability of the aggregation personnel number change rate.
Specifically, in order to obtain a normal behavior sample of the first monitoring blind area in a plurality of time zones, data mining is performed by analyzing the historical monitoring video data. Firstly, the monitoring videos in different time periods are extracted, and common activity types in blind areas, such as personnel passing, gathering or strolling, are identified and classified. Next, the frequency, duration and pattern of occurrence of the behavior in each time zone are recorded, and a conventional behavior database is built. Through the samples, a reference standard is provided for subsequent abnormal detection so as to effectively identify deviation from a conventional behavior mode and further improve the response capability to abnormal behaviors.
Feature extraction is performed based on the obtained regular behavior samples to generate aggregate features for a plurality of time zones. Firstly, analyzing behavior data in different time periods, and extracting the quantity of the aggregation personnel and the change trend of the aggregation personnel. For each time zone, a threshold number of people aggregated is calculated, i.e., the minimum number of people in the time zone that are considered aggregated. This threshold is set according to the average level and standard deviation of the historical data to ensure that routine behavior is accurately reflected. In addition, the rate of change of the aggregate personnel is analyzed to extract the stability characteristics. This includes calculating the rate of increase or decrease of the aggregate population over a specified period of time to identify normal fluctuations and abnormal spikes or decreases. Through the aggregation characteristics, a dynamic monitoring model is established, so that abnormal aggregation behaviors are effectively identified in real-time monitoring, and timely early warning is provided for campus safety.
The aggregation personnel change rate stabilization feature is used for measuring the stability and consistency of aggregation personnel change. This feature reflects the regularity of aggregation behavior by analyzing the rate of change of the number of aggregates in a specific period of time, records the number of aggregates at different points in time (e.g., every minute, hour), and calculates the rate of change, i.e., the increase or decrease in the number of people per unit time. 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 is indicated that the fluctuation of the number of people gathered in the time zone is small and the change is relatively stable, and conversely, if the standard deviation is large, it is indicated that the change of the number of people gathered is relatively severe and there is a risk of abnormal behavior. Furthermore, a threshold for the rate of change is established beyond which the rate of change will be considered unstable, indicating abnormal aggregation or personnel flow conditions. By monitoring the stable characteristic, potential safety hazards are better identified, and powerful data support is provided for campus management.
In one possible implementation manner, the first anomaly probability generation module 60 further includes:
the method comprises the steps of obtaining a first monitoring time zone of first monitoring video data, matching target standard aggregation characteristics in a first standard aggregation characteristic set according to the first monitoring time zone, comparing the first real-time aggregation characteristics with the matched target standard aggregation characteristics, establishing aggregation characteristic deviation, combining a monitoring object to conduct fusion of abnormal weights, and generating first abnormal probability.
Specifically, related monitoring video data is firstly extracted from the monitoring equipment, the process ensures that the acquired data is complete and is not tampered, timestamp information in the video data is analyzed to determine a specific recording time period, and metadata of a video file is extracted to realize the process. The time stamps are classified according to the set time ranges, so that a specific first monitoring time zone is divided, for example, a day is divided into different time periods of morning, afternoon, evening and the like. Finally, the identified monitored time zone is validated, ready for subsequent aggregate feature matching and abnormal behavior detection. This procedure ensures accurate grasp of the monitored time zone, which helps to promote the effectiveness of subsequent data analysis.
And according to the first monitoring time zone, analyzing the first standard aggregation characteristic set to match the target standard aggregation characteristic. Firstly, extracting real-time aggregation characteristics in the time zone, including the number of people currently aggregated, the activity type, the change trend and the like. These real-time features are then compared to historical patterns of behavior in the first set of standard aggregated features, looking for the most similar features. This matching process involves calculating the similarity between features using the euclidean distance algorithm to determine the degree of variance of the current aggregate features from the historical regular behavior. By this matching, the target standard aggregation feature is identified as a reference for subsequent anomaly detection. The process is not only helpful for defining the aggregation behavior of the current monitoring time zone, but also provides key data support for identifying potential anomalies, thereby improving the intelligent and response capability of the monitoring system.
And determining a first real-time aggregation characteristic and a target standard aggregation characteristic matched with the current time zone, and determining real-time aggregation characteristics such as the actual personnel number, the personnel change rate and the like at the current moment for the first monitoring blind area. And simultaneously, finding out a standard aggregation personnel quantity threshold value and a standard personnel change rate stable characteristic of the corresponding time zone from the first standard aggregation characteristic set. And then calculating the aggregate characteristic deviation, namely calculating the absolute value of the difference between the actual personnel number and the standard personnel number threshold value, and calculating the personnel change rate deviation which is the absolute value of the difference between the actual personnel change rate and the standard personnel change rate stable characteristic. And then, synthesizing the two deviations in a Euclidean distance mode to obtain an aggregate characteristic deviation value. For each monitoring object entering the first monitoring blind area, determining an abnormal weight according to the historical campus abnormal behavior record, counting the proportion of the times of occurrence of abnormal behaviors of each monitoring object in history to the total behavior record times as the abnormal weight of the monitoring object, calculating the sum of the abnormal weights of all the monitoring objects, dividing the abnormal weight of each monitoring object by the sum, and carrying out normalization processing. When the first anomaly probability is calculated, the importance coefficients of the aggregate characteristic deviation and the anomaly weight are set, the sum of the importance coefficients is 1, an adjustment factor is set for each monitored object, the first anomaly probability is determined according to factors such as the severity of historical anomaly behaviors, the first anomaly probability is equal to the aggregate characteristic deviation multiplied by the importance coefficient of the first anomaly probability, and the sum of the normalized anomaly weight and the adjustment factor of all the monitored objects multiplied by the importance coefficient of the anomaly weight.
In one possible implementation manner, the first anomaly probability generation module 60 further includes:
Initializing abnormal weights of all objects entering the first monitoring blind area in the first monitoring object time sequence to generate an initialized abnormal weight set, identifying initial abnormal probability based on the aggregation characteristic deviation, wherein the initial abnormal probability is in direct proportion to the aggregation characteristic deviation, and calculating the weighted influence of the initialized abnormal weight set on the initial abnormal probability to generate the first abnormal probability.
Specifically, all the objects entering the first monitoring blind area in the first monitoring object time sequence are initialized with abnormal weights, and at this stage, an initial abnormal weight is allocated according to each object entering the blind area. This weight is determined based on a number of basic factors such as the type of object (student, teacher, staff, etc.), the frequency of entering the blind zone, etc. And initializing the abnormal weights of all the objects to generate an initialized abnormal weight set.
Then, an initial anomaly probability is identified based on the aggregate feature deviation. The aggregate characteristic deviation is established by comparing the first real-time aggregate characteristic with the matching target standard aggregate characteristic, and reflects the difference between the current state and the normal state of the first monitoring blind area. The initial anomaly probability is proportional to the magnitude of the aggregate characteristic deviation, i.e., the greater the deviation, the higher the initial anomaly probability, because a greater deviation means that the current dead zone situation differs from the normal situation by a greater amount, and the likelihood of anomalies increases accordingly.
And finally, carrying out weighted influence calculation on the initial abnormal probability by using the initial abnormal weight set to generate a first abnormal probability. In this step, the anomaly weight of each object affects the initial anomaly probability. If the anomaly weight of an object is high, its contribution to the first anomaly probability is relatively large. By the weighting calculation, the influence degree of different objects on the abnormal condition of the dead zone can be reflected more accurately. For example, if some objects are historically often associated with an abnormal situation, their high anomaly weight may raise the first anomaly probability accordingly, thereby drawing more attention to the system. By initializing the abnormal weight of the object entering the first monitoring blind area, identifying the initial abnormal probability based on the aggregated characteristic deviation, and then carrying out weighted influence calculation, a more accurate and targeted first abnormal probability is generated, and an important basis is provided for campus abnormal behavior detection and early warning.
In one possible implementation manner, the first anomaly probability generation module 60 further includes:
The method comprises the steps of obtaining a historical campus abnormal behavior library, wherein any abnormal behavior sample in the historical campus abnormal behavior library comprises an abnormal person identification sample, counting first abnormal behavior frequency characteristics of any first person with abnormal behaviors based on the abnormal person identification sample, initializing abnormal weights of the first person based on first abnormal behavior frequency characteristic configuration to generate first abnormal weights, and adding the first abnormal weights to an initialized abnormal weight set when an object entering a first monitoring blind area comprises the first person.
Specifically, it is an important link to obtain a historical campus abnormal behavior library, where related information of various abnormal behaviors occurring in the past is stored. Any abnormal behavior sample comprises abnormal personnel identification samples which clearly indicate specific personnel participating in the abnormal behavior, and the historical campus abnormal behavior library is obtained to analyze and study the past abnormal behavior, so that possible abnormal conditions in the future can be better identified and predicted. For example, by analyzing the abnormal person identification sample, the person who has participated in the abnormal behaviors and the specific characteristics and modes of the abnormal behaviors are known, and an important reference basis is provided for subsequent abnormal behavior detection and early warning.
Counting the frequency of abnormal behaviors of the first person in detail, wherein the frequency and the frequency of abnormal behaviors of the first person are recorded, the distribution of the behaviors in different time periods is recorded, and the abnormal behavior patterns of the first person are known through analysis of the data, such as whether specific time rules exist, whether specific events or situations are related, and the like. The statistical analysis can provide important basis for subsequent abnormal behavior detection, judge the possibility of the first person appearing abnormal behavior again according to the first abnormal behavior frequency characteristic, and correspondingly adjust the attention degree and the early warning level of the first person. Meanwhile, the characteristics are also beneficial to the system to find potential abnormal behavior trend, and measures are taken in advance to prevent and intervene, so that the safety and stability of the campus are improved.
The process of initializing the abnormal weight is to assign a value reflecting the tendency of the abnormal behavior of the first person, which is the first abnormal weight. If the frequency of abnormal behaviors of the first person is higher, the corresponding first abnormal weight of the first person is correspondingly increased, which indicates that the person has a higher possibility of abnormal behaviors and needs to pay more attention. Conversely, if the first person's abnormal behavior is less frequent, then his first abnormal weight will be relatively less. The first abnormal weight generated in the mode plays an important role in subsequent abnormal probability calculation and early warning generation, and comprehensively considers the first abnormal weight of each person and other related factors to judge whether abnormal conditions exist in the campus more accurately and send early warning information timely so as to ensure the safety and order of the campus.
And when an object enters the first monitoring blind area, real-time monitoring and analysis are carried out. If an individual previously marked as a first person is included in the object entering the blind zone, a first outlier weight corresponding to the first person is added to the initialized outlier weight set. By adding the first abnormal weight to the set, the historical abnormal behavior condition of the person entering the blind area is more comprehensively considered, the first abnormal weight reflects the possibility or tendency degree of the abnormal behavior of the first person, and the first abnormal weight is added to the set, so that the abnormal degree of the current condition can be more accurately evaluated in subsequent abnormal probability calculation. For example, if a first person historically experiences abnormal behavior multiple times with a first abnormal weight that is higher, then when he enters the first surveillance blind zone, he is more concerned about this situation, as his presence may increase the risk of the blind zone experiencing an abnormal condition. By the mode, potential abnormal conditions can be more effectively identified and pre-warned, and the safety and management efficiency of the campus are improved.
In one possible implementation manner, the first anomaly probability generation module 60 further includes:
If the object entering the first monitoring blind area does not appear in the historical campus abnormal behavior library, initializing the corresponding abnormal weight to be zero.
Specifically, for an object entering the first monitoring blind area, a check is performed to determine whether it has occurred in the historical campus abnormal behavior library, and if some object does not occur in the library, this means that there is no record about the past abnormal behavior of the object. In this case, in order to accurately evaluate the abnormal situation, the abnormal weight of the object that does not appear in the history base is initialized. Because there is no history data supporting the possibility that the object has abnormal behavior, giving it zero weight can avoid misjudging the probability of abnormality. Therefore, when the first anomaly probability is calculated, the first anomaly probability is not disturbed by the objects without the history anomaly record, the objects with potential anomaly risks can be more accurately focused, and the accuracy and the reliability of anomaly behavior detection are improved.
In one possible implementation manner, the first anomaly probability generation module 60 further includes:
The method comprises the steps of obtaining non-zero weights in an initialization abnormal weight set, obtaining duty ratio coefficients of the non-zero weights, weighting initial abnormal probability after overlapping the non-zero weights when the duty ratio coefficients are larger than a preset duty ratio coefficient threshold value, and weighting the initial abnormal probability by a weight maximum value when the duty ratio coefficients are smaller than or equal to the preset duty ratio coefficient threshold value to generate the first abnormal probability.
Specifically, the initialization abnormal weight set is traversed, and elements with weight values different from zero are screened out. The non-zero weights represent that the monitoring object corresponding to the non-zero weights has abnormal behavior records in the historical campus abnormal behavior library, and the possibility of abnormal behaviors in the current monitoring scene is more accurately estimated by acquiring the non-zero weights, so that key information support is provided for subsequent abnormal probability calculation and early warning decision.
The duty cycle coefficients of the non-zero weights are calculated and compared to a preset duty cycle coefficient threshold. If the duty ratio coefficient is larger than the preset duty ratio coefficient threshold value, the proportion of the non-zero weight in the whole is larger. At this time, the overlapping operation is performed on these non-zero weights, and the result after the overlapping will be used to weight the initial anomaly probability. In this way, the influence of a plurality of objects with abnormal weights on the abnormal probability is fully considered, so that the calculated first abnormal probability can more accurately reflect the actual situation. And when the duty cycle is less than or equal to the preset duty cycle threshold, this means that the proportion of the non-zero weight in the whole is small, in which case the initial anomaly probability is weighted with the maximum value of the weight. The maximum weight value is the maximum weight value appearing in the initialized abnormal weight set, and the purpose of doing so is to still give certain adjustment to the initial abnormal probability when the non-zero weight proportion is smaller so as to ensure the accuracy of the first abnormal probability. And the first abnormal probability is reasonably generated by selecting different weighting modes according to the size of the duty ratio coefficient, so that the accuracy and the reliability of campus abnormal behavior detection are improved.
It should be noted that the 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 foregoing description has been directed to specific embodiments of this specification. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
The foregoing description of the preferred embodiments of the application is not intended to limit the application to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the application are intended to be included within the scope of the application.
The specification and figures are merely exemplary illustrations of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents that fall within the scope of the application. It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the scope of the application. Thus, the present application is intended to include such modifications and alterations insofar as they come within the scope of the application or the equivalents thereof.