A video surveillance method for abnormal behavior of community personnel based on human factors engineering

Abstract

The present invention relates to the field of community video surveillance technology, and discloses a method for video surveillance of abnormal behavior of community personnel based on human factors engineering. The method first divides community surveillance video data based on preset activity areas to obtain monitoring data of each sub-area, including personnel trajectory and environmental interaction data; then determines the abnormality level of each sub-area based on the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data; then calculates key monitoring nodes based on the abnormality level, and determines the scanning order and scanning duration according to dynamic path adjustment; then generates observation anchor points and redundant anchor points at the sub-area boundaries; and finally, associates and integrates them to generate a monitoring path. The method of the present invention improves the accuracy, flexibility, and environmental adaptability of community abnormal behavior monitoring through multi-dimensional data processing and dynamic optimization, and is suitable for community safety management.

Description

Technical Field

[0001] The present invention relates to the technical field of community video surveillance, and in particular to a method for video surveillance of abnormal behavior of community personnel based on human factors engineering. Background Art

[0002] With the acceleration of urbanization, the composition of community personnel is becoming increasingly complex. The traditional fixed-point, single-perspective video surveillance model has gradually exposed many limitations and is unable to meet the needs of modern community security management for accurate identification and dynamic monitoring of abnormal behavior of personnel.

[0003] In terms of monitoring scope and accuracy, traditional methods typically monitor communities as a whole, lacking a detailed spatial breakdown. This results in mixed monitoring data, making it difficult to quickly pinpoint anomalies in specific areas. For example, in scenarios like crowded squares or narrow corridors, behavioral patterns vary significantly. However, traditional monitoring methods cannot tailor monitoring to the characteristics of different areas, and are prone to missing anomalous behavior in localized areas, such as small groups gathering or individuals staying for extended periods of time.

[0004] When it comes to identifying abnormal behavior, existing technologies mostly rely on extracting behavioral features with fixed thresholds, failing to fully consider the interplay between the environment and human behavior in human factors engineering. For example, they identify anomalies solely based on movement speed or dwell time, ignoring the impact of environmental factors (such as building layout and facility distribution) on human behavior. In communities, certain areas (such as blind spots or facility concentrations) may exhibit brief periods of abnormal human interaction due to environmental characteristics. Traditional methods struggle to distinguish between normal and truly abnormal behavior, leading to false positives and missed alerts.

[0005] Traditional methods for planning monitoring paths often rely on fixed scanning sequences and durations, lacking dynamic adjustment mechanisms. When unexpected incidents occur within a community, monitoring paths cannot be quickly optimized based on real-time data, resulting in delayed monitoring of key nodes. For example, if a sudden gathering of people occurs in a certain area, fixed scanning sequences and durations may not be able to promptly include that area in key monitoring, delaying the detection and resolution of abnormal behavior.

[0006] Furthermore, existing technologies are insufficiently adaptable to the dynamics of community environments. Changes to building layouts within a community (such as the addition of temporary facilities or road improvements) can alter regular paths and gathering areas for people. However, traditional monitoring systems struggle to update their monitoring range and parameters in real time, leading to gaps in coverage. For example, a newly added express delivery point may become a new gathering area. If traditional systems fail to adjust their monitoring strategies in a timely manner, they may be unable to effectively detect abnormal behavior in that area.

[0007] When it comes to data fusion and processing, traditional methods typically analyze either individual trajectory data or environmental data in isolation, failing to fully exploit the correlations between them. For example, a sudden change in a person's trajectory might be associated with an obstacle or an unusual event in the environment. However, traditional methods are unable to quickly identify such potential anomalies through correlation analysis, limiting their ability to comprehensively identify complex abnormal behaviors. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for video monitoring of abnormal behavior of community personnel based on human factors engineering to solve the problems raised in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A method for video monitoring abnormal behavior of community personnel based on human factors engineering, the method comprising:

[0011] The community surveillance video data is spatially divided based on the preset activity area range to obtain the surveillance data of each sub-area, wherein the surveillance data of each sub-area includes personnel trajectory data and environmental interaction data;

[0012] Determine the abnormality level of each sub-area based on the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data in each sub-area;

[0013] Calculate the key monitoring nodes of each sub-area based on the anomaly level, personnel trajectory data, and environmental interaction data of each sub-area, and determine the scanning order and scanning duration of the key monitoring nodes based on dynamic path adjustment;

[0014] Generate observation anchor points for each sub-area at the boundary of each sub-area based on the preset observation angle, key monitoring nodes of each sub-area, and spatial orientation information, and generate redundant anchor points for each sub-area at the boundary of each sub-area based on the preset redundant distance, the key monitoring nodes, and spatial orientation information;

[0015] The key monitoring nodes, observation anchor points and redundant anchor points are associated and integrated based on the scanning order and scanning duration of the key monitoring nodes to generate a video monitoring path for the community.

[0016] Furthermore, the community surveillance video data is spatially divided based on the preset activity area range to obtain the surveillance data of each sub-area, including:

[0017] For the current sub-region, the division boundary of the current sub-region is calculated according to the geometric center coordinates of the current sub-region, the boundary coordinates of the adjacent sub-regions and the preset spatial overlap threshold;

[0018] Extracting surveillance video data of consecutive frames based on the division boundary of the current sub-region and a preset time window;

[0019] The movement trajectories of personnel and the environmental interaction events that conform to the division boundary in the monitoring video data of the continuous frames are marked as the monitoring data of the current sub-area.

[0020] Furthermore, the abnormality level of each sub-area is determined based on the movement frequency of the personnel trajectory data and the interaction intensity of the environment interaction data of each sub-area, including:

[0021] For the current sub-region, when the number of movement per unit time of the personnel trajectory data exceeds a preset frequency threshold, the current sub-region is marked as a first-level abnormal region;

[0022] When the number of movements per unit time does not exceed the preset frequency threshold, performing spatiotemporal clustering on the environmental interaction data based on a preset clustering radius to obtain interaction event clusters;

[0023] The difference between the duration of each interaction event cluster and the preset benchmark duration is recorded as the duration difference, and the interaction anomaly index is calculated based on the duration difference;

[0024] If the interaction anomaly index exceeds a preset anomaly threshold, marking the current sub-region as a secondary anomaly region;

[0025] If the preset abnormal threshold is not exceeded, the current sub-area is marked as a regular monitoring area.

[0026] Furthermore, the monitoring data of each sub-area also includes wearing feature data and group gathering data. The key monitoring nodes include high-risk behavior nodes, retention nodes, and group gathering nodes. The key monitoring nodes of each sub-area are calculated based on the abnormality level, personnel trajectory data, and environmental interaction data of each sub-area, including:

[0027] For the current sub-region, the group aggregation data is screened based on a preset density threshold to obtain a high-density aggregation region;

[0028] Calculate the gathering risk coefficient based on the uniformity of personnel distribution and consistency of movement direction in the high-density gathering area;

[0029] If the aggregation risk coefficient exceeds a preset risk threshold, the geometric center of the high-density aggregation area is determined as the group aggregation node;

[0030] The coordinate points whose stay time exceeds a preset time length are extracted from the personnel trajectory data, and are marked as the stay nodes in combination with the interaction type of the environmental interaction data.

[0031] Furthermore, after calculating the key monitoring nodes, the method further includes:

[0032] generating a candidate set of high-risk behavior nodes based on the occurrence frequency and spatial distribution density of abnormal wearing identifiers in the wearing feature data;

[0033] Performing redundancy filtering based on the spatiotemporal correlation of each candidate node in the candidate set of high-risk behavior nodes, and retaining candidate nodes that meet a preset correlation strength;

[0034] The spatial coordinates of the retained candidate nodes that meet the preset association strength are vector-matched with the movement direction of the personnel trajectory data, and the candidate nodes with a direction deviation greater than a preset angle are screened out as the final high-risk behavior nodes.

[0035] Furthermore, the dynamic path adjustment includes:

[0036] Arranging the clustered nodes in descending order of clustering risk coefficients to generate a first scanning sequence;

[0037] Arrange the stranded nodes in descending order according to the ratio of the stay time to the preset time length to generate a second scanning sequence;

[0038] The total scanning time is set, and the two scanning sequences are allocated the total scanning time according to the time weight. The time weights of the first scanning sequence and the second scanning sequence are dynamically adjusted based on the average value of the difference between the direction deviation of all high-risk behavior nodes and the preset angle. The scanning time of the first scanning sequence and the second scanning sequence is updated, and the monitoring scan is performed with the updated scanning time of the two scanning sequences. The scanning sequence with a large time weight is scanned first to obtain the scanning order and scanning time of the key monitoring nodes.

[0039] Furthermore, the dynamic path adjustment further includes:

[0040] Real-time monitoring of the direction deviation change rate of the high-risk behavior node, and triggering a path backtracking mechanism when the direction deviation change rate exceeds a preset fluctuation threshold;

[0041] Extracting the scanning records of the group gathering nodes and the stranded nodes in the previous period based on the path backtracking mechanism, and recalculating the duration weight;

[0042] The recalculated duration weight is interpolated and fused with the duration weights of the current first scanning sequence and the second scanning sequence to generate a new anti-interference weight, and the scanning order and scanning duration are re-determined with the new anti-interference weight.

[0043] Furthermore, the calculation of the interaction anomaly index further includes:

[0044] Extracting the body movement amplitude data of the participants in each interaction event cluster, and matching abnormal movement patterns based on a preset movement standard library;

[0045] The successfully matched abnormal action patterns are weighted and summed according to the danger level to generate the action risk coefficient;

[0046] The action risk coefficient and the duration difference are linearly fused to output a corrected value of the interaction anomaly index.

[0047] Furthermore, after generating the video surveillance path, the following steps are also included:

[0048] Real-time detection of the offset distance between each key monitoring node in the video monitoring path and the real-time position of the personnel;

[0049] If the offset distance exceeds a preset fault tolerance threshold, the spatial coordinates of the observation anchor point are dynamically updated according to the relative orientation between the real-time position of the person and the preset activity area;

[0050] The generation position of the redundant anchor point is recalculated based on the updated observation anchor point, and the video monitoring path is iteratively optimized.

[0051] Furthermore, the preset activity area range is dynamically adjusted, including:

[0052] Obtain real-time data on building layout changes within the community and generate a virtual space grid based on the changed topological structure;

[0053] Dynamically expand or contract the boundary threshold of the activity area according to the difference between the historical data and the real-time data of the density distribution of people in the virtual space grid;

[0054] The boundary threshold is superimposed on the preset minimum safety distance to generate an updated space division parameter.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] In terms of spatial division and data processing, community surveillance video data is spatially partitioned using pre-set activity zones. Sub-region boundaries are determined by combining geometric center coordinates, adjacent boundary coordinates, and a pre-set spatial overlap threshold, ensuring the independence and relevance of surveillance data within each sub-region. Consecutive frames of surveillance video data are extracted based on the boundaries and time windows, and human movement trajectories and environmental interaction events that fall within these boundaries are marked. This enables refined monitoring of community spaces, avoids the data clutter associated with traditional monolithic surveillance models, and provides a precise data foundation for subsequent analysis of abnormal behavior.

[0057] Abnormality level assessment combines the movement frequency of individual trajectory data with the interaction intensity of environmental interaction data. By setting frequency thresholds and performing spatiotemporal cluster analysis, it can accurately distinguish between primary and secondary abnormal areas, as well as regular monitoring areas. Furthermore, the introduction of wearable feature data and group aggregation data further refines the calculation of key monitoring nodes (high-risk behavior nodes, detention nodes, and group aggregation nodes). For example, high-density aggregation areas are screened using a preset density threshold, and the aggregation risk coefficient is calculated based on the uniformity of individual distribution and the consistency of movement directions. This ensures comprehensive monitoring of different types of abnormal behavior, improves the accuracy and comprehensiveness of abnormal behavior identification, and reduces false positives and missed alerts.

[0058] The scanning order and scanning duration of key monitoring nodes are generated based on dynamic path adjustment. Group gathering nodes are sorted in descending order by the aggregation risk coefficient, and the detention nodes are sorted in descending order by the ratio of the residence time to the preset duration. The duration weight is dynamically adjusted according to the directional deviation of high-risk behavior nodes, and the nodes with large duration weights are scanned first. This dynamic adjustment mechanism enables the monitoring system to flexibly optimize the scanning order and scanning duration according to real-time abnormal situations, ensuring that high-risk areas and key nodes are monitored first, improving the response speed and processing efficiency of abnormal events. In addition, a path backtracking mechanism is introduced. When the rate of change of the directional deviation of high-risk behavior nodes exceeds the preset fluctuation threshold, the duration weight is recalculated and a new weight is generated to resist interference, which enhances the stability and anti-interference ability of the monitoring path.

[0059] The generation mechanism for observation anchors and redundant anchors combines preset observation angles, redundant distances, and spatial orientation information to generate anchors at sub-area boundaries and integrate them into a video surveillance path. Simultaneously, by detecting the offset distance between key monitoring nodes and the actual location of personnel, the positions of observation anchors and redundant anchors are dynamically updated, and the surveillance path is iteratively optimized. This allows the surveillance system to adapt to personnel position changes in real time, ensuring the accuracy and continuity of the surveillance image and avoiding blind spots caused by personnel movement.

[0060] The dynamic adjustment of pre-set activity zones is based on data on changes in community building layouts and population density. By generating a virtual spatial grid and adjusting boundary thresholds, the system adapts to the dynamic changes in the community environment. This dynamic adjustment ensures that the monitoring range always covers the actual activity areas of residents, avoiding monitoring loopholes caused by environmental changes and improving the environmental adaptability and long-term effectiveness of the monitoring system.

[0061] The correction calculation of the interaction anomaly index introduces limb movement amplitude data and a preset movement standard library. By matching abnormal movement patterns and weightedly calculating the movement risk coefficient, and linearly fusing it with the duration difference to generate a correction value, the accuracy of identifying potential abnormal behaviors in environmental interaction data is further improved, enabling the system to more accurately judge the dangerousness of interactive behaviors, providing a more reliable basis for early warning of abnormal behaviors. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of the process of the method for video monitoring abnormal behavior of community personnel based on human factors engineering of the present invention;

[0063] Figure 2 Schematic diagram of the spatial partitioning of community surveillance video data;

[0064] Figure 3 Schematic diagram of calculation for key monitoring nodes (group gathering nodes, retention nodes);

[0065] Figure 4 Schematic diagram of the dynamic path adjustment process. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] See also Figure 1-Figure 4 The present invention relates to a method for video monitoring abnormal behavior of community personnel based on human factors engineering, and the specific implementation steps are as follows:

[0068] By spatially segmenting community surveillance video data based on pre-set activity zones, we generate surveillance data for each sub-area, including both occupant trajectory data and environmental interaction data. This structured segmentation of the community's physical space enables refined monitoring of occupant activity in different areas.

[0069] Determine the anomaly level of each sub-area based on the movement frequency of personnel trajectory data and the interaction intensity of environmental interaction data. Identify areas requiring special attention through quantitative analysis of personnel movement characteristics and environmental interaction patterns.

[0070] Based on the anomaly level, personnel trajectory data, and environmental interaction data of each sub-area, the key monitoring nodes of each sub-area are calculated. The scanning order and scanning duration of key monitoring nodes are determined based on dynamic path adjustment. By establishing a multi-level key monitoring node system, the optimal configuration of monitoring resources is achieved.

[0071] Based on the preset observation angles, key monitoring nodes, and spatial orientation information for each sub-area, observation anchor points are generated at the boundaries of each sub-area. Simultaneously, based on the preset redundant distances, key monitoring nodes, and spatial orientation information, redundant anchor points are generated at the boundaries of each sub-area. By setting anchor points with different functions, the comprehensiveness and reliability of monitoring coverage are improved.

[0072] Based on the scanning order and duration of key monitoring nodes, key monitoring nodes, observation anchor points, and redundant anchor points are associated and integrated to generate a community video surveillance path. By building a dynamically adjustable video surveillance path, it is possible to efficiently capture and track abnormal behavior.

[0073] Example 1:

[0074] Based on the above overall plan, the specific implementation of space division is as follows:

[0075] For the current sub-region, its geometric center coordinates must first be determined. These coordinates can be obtained using the community's spatial coordinate system. For example, based on the Gaussian plane coordinate system or the Universal Time Coordinate System (UTM) coordinate system, the sub-region is abstracted into a geometric shape (such as a rectangle or circle) and its geometric center coordinates are calculated. The boundary coordinates of adjacent sub-regions are determined by the coordinates of the boundary vertices of the previously divided geometric shapes of other sub-regions. The geometric shapes of each sub-region can be divided into irregular polygons based on the actual community layout (such as road and building distribution) to adapt to real-world surveillance scenarios. The preset spatial overlap threshold is a configurable distance parameter, for example, set to 1-5 meters. It ensures that there is overlap in surveillance coverage between adjacent sub-regions to avoid blind spots caused by camera field of view limitations or installation position deviations. When calculating the boundary, the geometric center coordinates of the current sub-region are used as a reference. The boundary coordinates of the adjacent sub-regions are combined with the preset spatial overlap threshold. Geometric operations (such as vector calculations and polygon intersection) are used to determine the boundary vertex coordinates of the current sub-region, forming a closed polygonal boundary. For example, if the boundary of the adjacent sub-region is D meters away from the geometric center of the current sub-region and the preset overlap threshold is S meters, the boundary of the current sub-region can be extended by S meters in the adjacent direction so that the two sub-regions overlap by S meters at the junction.

[0076] After completing boundary demarcation, continuous frames of surveillance video data are extracted based on the current sub-region boundary and a preset time window. The preset time window is a continuous time period for analysis and can be set to 5 minutes, 10 minutes, or other timeframes based on monitoring requirements. The system segments the video stream based on timestamps and extracts all frames within that timeframe. During the extraction process, video analysis techniques are used to process each frame. Object detection algorithms (such as YOLO and SSD) are used to identify people within the frames, and object tracking algorithms (such as DeepSORT) are used to generate person movement trajectories. For each trajectory, the system determines in real time whether the trajectory point lies within the current sub-region boundary. This is achieved using a point-to-polygon containment detection algorithm (such as the ray method): a ray is drawn from the trajectory point in any direction and the number of intersections between the ray and the sub-region boundary polygon is calculated. If the number of intersections is odd, the point is considered to be within the polygon; otherwise, it is outside. Person movement trajectory data that fits within the boundary is recorded, including parameters such as the trajectory point coordinates, timestamp, and movement speed.

[0077] At the same time, the system detects and marks environmental interaction events in consecutive frames. Environmental interaction events include interactions between people and objects or facilities in the community environment, such as pushing access control, picking up objects on the ground, and approaching dangerous areas (such as sinks and electrical boxes). Through image recognition technology and a preset interactive behavior rule library, the human actions and environmental elements in the video frames are matched and identified. For example, when a person's hand is detected approaching an object on the ground and making a grasping action, it is marked as an "object pickup" event; when a person continues to approach the boundary of a dangerous area where a sensor is installed, it is marked as a "dangerous area approach" event. The time, location, and type of these interaction events will be recorded as environmental interaction data.

[0078] In practical applications, sub-areas can be differentiated based on the functional areas of a community. For example, main roads with dense traffic flow can be divided into smaller sub-areas (such as polygons with a side length of 10 meters) to improve monitoring accuracy. Relatively open areas such as green belts and parking lots can be divided into larger sub-areas (such as polygons with a side length of 20 meters) to reduce computing resource consumption. The calculation of the division boundary must take into account the camera's installation location and viewing angle range to ensure that each sub-area is covered by the effective monitoring range of at least one camera. For example, if a camera has a monitoring angle of 120 degrees horizontally and 60 degrees vertically, and an effective monitoring distance of 30 meters, the sub-area boundary divided within the viewing angle of the camera, centered on the camera, should be within the viewing angle coverage area within a distance of 30 meters.

[0079] In addition, the system needs to regularly calibrate the boundaries to cope with changes in the monitoring range caused by changes in the community environment (such as the installation of temporary obstacles and road construction). During the calibration process, the boundary coordinates of adjacent sub-areas are updated manually or automatically, and the boundaries of the current sub-area are recalculated to ensure the accuracy of the spatial division. When extracting monitoring data, for the trajectories of people moving across sub-areas, the system will automatically segment the trajectory data according to the time sequence, and attribute the trajectory points within the boundary of the current sub-area in different time periods to the trajectory data of the people in the current sub-area, and the trajectory points outside the boundary are attributed to the data of the adjacent sub-area, so as to achieve seamless connection and accurate classification of cross-regional trajectories.

[0080] The labeling of environmental interaction data requires filtering based on contextual information to avoid false positives. For example, in an "item pickup" event, if the person picks up a common item (such as a schoolbag or water bottle) and leaves normally after picking it up, it may be considered normal behavior. However, if the item picked up is a suspicious package and the person remains for an extended period or moves to a secluded area after picking it up, it will be flagged as a high-risk interaction event. By establishing a behavioral pattern knowledge base, the system classifies different types of interaction events by risk level, providing multi-dimensional data support for subsequent determination of anomaly levels.

[0081] In terms of data storage, monitoring data from each sub-area is stored in a time series format. Personnel trajectory data is indexed by trajectory ID, with each trajectory consisting of a series of coordinate points and corresponding timestamps. Environmental interaction data is indexed by event ID, with each event record containing the event type, occurrence time, location coordinates, and associated video frame indexes. This storage structure facilitates subsequent data retrieval, analysis, and visualization. For example, the trajectory playback function can be used to replay the movement paths of personnel within a sub-area, while the event statistics function can be used to analyze the frequency and spatial distribution of specific types of interaction events.

[0082] Example 2:

[0083] Regarding the method for determining the abnormality level, the specific implementation process is as follows:

[0084] For the current sub-area, the movement frequency in the person trajectory data is first quantified and analyzed. Person trajectory data contains the number of movements per unit time, calculated from the time interval between trajectory points. For example, if the preset time window is 10 minutes and the trajectory data contains N trajectory points, the number of movements per unit time is (N-1) / 10 (times / minute). The preset frequency threshold is an empirical parameter set based on the statistical value of the movement frequency of people in normal community activities, for example, 5 times / minute (indicating an average of one movement every 12 seconds). When the number of movements per unit time in the person trajectory data for the current sub-area exceeds the preset frequency threshold, it indicates abnormal movement, possibly involving pursuit or escape behavior. In this case, the current sub-area is marked as a Level 1 abnormal area, triggering the highest level of surveillance response, such as increasing the surveillance scanning frequency in the area or initiating multi-camera coordinated tracking.

[0085] If the number of movements per unit time does not exceed the preset frequency threshold, the process of spatiotemporal clustering analysis of the environmental interaction data begins. Spatiotemporal clustering analysis processes event points in the environmental interaction data based on a preset cluster radius. The preset cluster radius is a combination of a spatial distance parameter (i.e., a preset spatial distance, such as 3 meters) and a time interval parameter (i.e., a preset time interval, such as 2 minutes). It is used to determine whether events belong to the same cluster: if the spatial distance between the occurrence locations of two interaction events is less than or equal to the preset spatial radius, and the time difference between them is less than or equal to the preset time interval, they are considered to belong to the same interaction event cluster. Using a density clustering algorithm such as DBSCAN, all event points in the environmental interaction data are clustered to obtain several interaction event clusters, each representing a group of interaction behaviors that occurred within a similar spatiotemporal range.

[0086] The difference between the duration of each interaction event cluster and a preset benchmark duration is recorded as the duration difference. The interaction anomaly index is calculated based on the duration difference. The preset benchmark duration is a reference value set based on the average duration of normal interaction behavior, for example, 5 minutes. For each interaction event cluster, its duration is the time difference between the last and first events in the cluster. If a cluster lasts T minutes, the duration difference is |T-5| minutes. This duration difference reflects the degree to which the duration of the interaction behavior in that cluster deviates from the normal level; a larger difference indicates more abnormal interaction behavior. The interaction anomaly index is calculated by averaging the duration differences of all interaction event clusters. The preset anomaly threshold is a critical value derived from historical data statistics, for example, set to 2 minutes. When the average duration difference of all interaction event clusters (i.e., the interaction anomaly index) exceeds the preset anomaly threshold, it indicates that the overall interaction behavior in the sub-area is abnormal and the sub-area is marked as a Level 2 anomaly area, requiring increased monitoring attention. If it does not exceed the threshold, the sub-area is marked as a regular monitoring area and scanned at the standard frequency.

[0087] During the analysis process, clothing feature data and group aggregation data in the monitoring data provide additional basis for determining abnormality levels. Wearing feature data is extracted using image recognition technology and includes information such as the color and style of clothing worn by individuals, and whether they are wearing special identification (such as hard hats or reflective vests). For example, detecting someone wearing a hard hat in a non-construction area, or someone wearing emergency rescue clothing during normal hours, can be considered an abnormal wear feature. The system records the number of occurrences and location distribution of these abnormal wear features as auxiliary judgment factors. Group aggregation data is generated using a density detection algorithm. This algorithm calculates the number of people per unit area based on the distribution of people in the video frames. When the density exceeds a preset density threshold (e.g., 3 people / square meter), it is determined to be a high-density gathering area.

[0088] For the determination of first-level abnormal areas, in addition to movement frequency, cross-validation can also be performed in conjunction with wearable feature data. For example, if the movement frequency of people in a sub-area exceeds a preset frequency threshold and multiple abnormal wearable identifications are detected at the same time, this further confirms the presence of high-risk abnormal behavior in the area, increasing the credibility of the abnormality level. For second-level abnormal areas, if multiple interactions involving people wearing abnormal wearable devices are found in an interaction event cluster, the calculation weight of each interaction event cluster when calculating the interaction anomaly index will be adjusted to increase the impact of such events on the abnormality level.

[0089] During spatiotemporal clustering, cross-subregion interaction events must be processed. If the event points of a cluster of interaction events are distributed across multiple subregions, the system automatically identifies all subregions involved in the cluster and includes the relevant data for each subregion in the anomaly level calculation. For example, for a cluster of events spanning two subregions, the difference in duration and the number of abnormal wear tags will be reflected in the interaction anomaly index calculation for each subregion, avoiding event fragmentation due to regional divisions.

[0090] In addition, the system supports dynamic adjustment of parameters such as preset frequency thresholds, cluster radius, and baseline duration. For example, during peak community activity periods (such as morning and evening commutes), when normal human movement is high, the preset frequency threshold can be automatically increased. During nighttime, when human activity decreases, the preset frequency threshold can be lowered to increase sensitivity to abnormal behavior. Dynamic parameter adjustment rules are generated based on historical data statistical models. By analyzing human behavior patterns at different times and under different weather conditions, they automatically optimize parameter values ​​and improve the adaptability of abnormality level judgment.

[0091] In terms of data processing efficiency, a distributed computing framework is used to parallelize the processing of human trajectory data and environmental interaction data for large-scale surveillance video data. For example, video data from different sub-areas is distributed to different computing nodes for trajectory extraction and event detection. Message queues are used to synchronize data between computing nodes and aggregate clustering results, ensuring rapid, real-time determination of anomaly levels.

[0092] Abnormality level marking results are displayed in real time through a visual interface. The community monitoring center can view the abnormality level status of each sub-area (for example, level 1 abnormal areas are marked in red, level 2 in yellow, and normal areas in green), and automatically trigger corresponding monitoring strategies based on the level. For example, level 1 abnormal areas trigger audible and visual alarms and automatically save recent video recordings of the area; level 2 abnormal areas initiate slow-motion playback and key frame marking, allowing monitoring personnel to quickly locate abnormal events.

[0093] Example 3:

[0094] The specific implementation methods for calculating key monitoring nodes and dynamic path adjustment are as follows:

[0095] For the current sub-area, the determination of group gathering nodes requires two stages: data screening and risk assessment. First, the group gathering data is screened based on the preset density threshold. The preset density threshold is set based on the statistical value of the population density of daily activities in the community. For example, it is set to 2.5 people / square meter. When the population density of a certain area in the sub-area exceeds the preset density threshold, it is determined to be a high-density gathering area. The population density is calculated based on the personnel detection results in the video frame. The specific method is: divide the sub-area into several grid cells, each with an area of ​​S2 square meters. Count the number of people in each cell N2, and then the population density of the unit is N2 / S2 (people / square meter). By traversing all grid cells, the areas with density exceeding the preset density threshold are screened as high-density gathering areas.

[0096] The cluster risk coefficient is calculated for high-density cluster areas. This coefficient is derived by combining two indicators: population distribution uniformity and movement direction consistency. Population distribution uniformity reflects the degree of spatial dispersion of the clustered population. It is calculated as the ratio of the standard deviation of the population density of each grid cell within the area to the mean density. A larger ratio indicates a more uneven distribution and a higher potential risk. Movement direction consistency is measured by calculating the mean cosine similarity of the movement direction vectors of each person in the cluster. A higher similarity indicates a more consistent movement direction (e.g., collectively moving in a certain direction), which may indicate an abnormal event (e.g., panic evacuation). The cluster risk coefficient combines these two indicators using a linear weighting method. The weights can be set based on the characteristics of risk events in historical data, for example, 40% for population distribution uniformity and 60% for movement direction consistency. If the cluster risk coefficient exceeds a preset risk threshold (e.g., 0.6), the geometric center of the high-density cluster area is identified as the cluster node, which is then included in the monitoring path as a key monitoring point.

[0097] The extraction of detention nodes is based on the dwell time in the personnel trajectory data. The system traverses all personnel trajectory points, and for each trajectory point, calculates the time interval from that point to the next trajectory point. If the time interval exceeds the preset time (such as 30 minutes), the trajectory point is determined to be a detention point. The preset time can be set according to the habit of staying in normal community activities. For example, the normal stay time allowed in the square rest area is 20 minutes, so the preset time is set to 30 minutes to distinguish abnormal stays. After extracting all coordinate points whose stay time exceeds the preset time, they are marked in combination with the interaction type of the environmental interaction data: if there are frequent environmental interaction events near the stay point (such as multiple operations of public facilities), it may be a normal stay; if the stay point is located in a remote area and there is no obvious interaction behavior, it is marked as a detention node, indicating that there may be abnormal wandering or safety hazards.

[0098] The generation of high-risk behavior nodes includes three steps: candidate set generation, redundancy filtering, and direction matching. First, a candidate set is generated based on the frequency and spatial distribution density of abnormal wear marks in the wear feature data. Abnormal wear marks include but are not limited to work clothes worn during non-working hours, clothing that obscures the face, and carrying suspicious items. The system detects the wear features of people in video frames through image recognition models (such as FasterR-CNN). When the number of abnormal wear marks in the same area per unit time exceeds a set threshold (such as 5 times / hour), or the spatial distribution of abnormal wear marks shows clustering (such as 3 cases within a 10-square-meter area), the coordinate points of the area are added to the candidate set.

[0099] The redundancy filtering phase eliminates duplicate or irrelevant nodes by analyzing the spatiotemporal correlations of each candidate node in the candidate set of high-risk behavior nodes. This spatiotemporal correlation calculation incorporates dual constraints: if the time difference between two candidate nodes is less than 15 minutes and the spatial distance is less than 5 meters, they are considered to be associated with the same high-risk behavior event, and only one candidate node (e.g., the earliest one) is retained as a representative. This step reduces redundant nodes resulting from multiple detections of the same abnormal behavior, improving the effectiveness of the node set.

[0100] Finally, the spatial coordinates of the retained candidate nodes that meet the preset correlation strength are vector-matched with the movement direction of the person's trajectory data. Specifically, for each candidate node, the person's movement direction vector at its location is obtained (calculated using the front and back coordinates of the trajectory point). The angle between this vector and the direction vector from the candidate node to the geometric center of the sub-area is calculated as the direction deviation. If the angle is greater than a preset angle (such as 60 degrees), it indicates that the person's movement direction has significantly deviated from the normal activity direction of the sub-area, indicating that there may be deliberate avoidance of monitoring or abnormal movement paths. Such nodes are screened as the final high-risk behavior nodes.

[0101] Dynamic path adjustment optimizes monitoring resource allocation through multi-sequence priority management. First, cluster nodes are sorted in descending order of risk to create a first scanning sequence, ensuring that high-risk cluster areas are monitored first. For example, a node with a risk factor of 0.8 is ranked before a node with a risk factor of 0.7. The monitoring system then scans each cluster node in this order, ensuring coverage of high-risk areas in the shortest possible time.

[0102] The second scanning sequence is generated by sorting the stranded nodes in descending order based on the ratio of their dwell time to the preset duration. A smaller ratio indicates a shorter dwell time (e.g., a dwell time of 40 minutes and a preset duration of 30 minutes results in a ratio of 1.33; a dwell time of 50 minutes results in a ratio of 1.67). After sorting in descending order, stranded nodes with longer dwell times are scanned first, allowing for timely detection of potential risks associated with prolonged dwellings.

[0103] The total scanning time is set, and the two scanning sequences are allocated the total scanning time according to the time weight. The time weights of the first scanning sequence and the second scanning sequence are dynamically adjusted based on the average value of the difference between the direction deviation of all high-risk behavior nodes and the preset angle. The scanning time of the first scanning sequence and the second scanning sequence is updated, and the monitoring scan is performed with the updated scanning time of the two scanning sequences. The scanning sequence with a large time weight is scanned first to obtain the scanning order and scanning time of the key monitoring nodes.

[0104] The directional deviation of high-risk behavior nodes is used to dynamically adjust the duration weights of the first two scanning sequences. The specific method is to calculate the average difference between the directional deviation of all high-risk behavior nodes and the preset angle. If the average value is greater than 10 degrees, it indicates that there are many abnormal directional behaviors in the current sub-area. The duration weight of the first scanning sequence (crowd gathering node) needs to be increased (for example, from the default 50% to 60%) to address possible group anomalies. If the average value is less than or equal to 10 degrees, the duration weight of the first scanning sequence is maintained or reduced, and the duration weight of the second scanning sequence is increased to focus on individual detention risks. By adjusting the duration weights in real time, dynamic responses to different types of anomalies are achieved.

[0105] At the system implementation level, the calculation of key monitoring nodes and the generation of scanning sequences and durations are accomplished through a real-time data processing module. This module receives monitoring data from each sub-area and uses multi-threaded parallel computing to calculate parameters such as the aggregation risk factor and directional deviation. It also utilizes a priority queue data structure to manage the scanning sequence for each type of key monitoring node. Once generated, the monitoring path is transmitted over the network to the camera control system, which drives the camera to sequentially scan each key monitoring node, simultaneously recording the scan time and video data to form a complete monitoring log.

[0106] The system also supports manual intervention in the scanning sequence. For example, if a monitoring operator discovers a sudden anomaly in a certain area through the visual interface, they can manually adjust the priority of nodes in that area to force them to be scanned earlier. Manual intervention is achieved through an event-triggered mechanism, ensuring a rapid response in emergency situations.

[0107] Example 4:

[0108] Regarding anchor point generation, monitoring path update, and dynamic adjustment of activity areas, the specific implementation is as follows:

[0109] When generating observation anchor points, the preset observation angle, the location of key monitoring nodes, and spatial orientation information must be comprehensively considered. The preset observation angle is a physical parameter of the camera or a software-configurable parameter. For example, the optimal observation angle for a certain camera model is 110 degrees horizontally and 70 degrees vertically. The system uses this angle to determine the sector area covered by the camera. For each key monitoring node (such as a gathering node or a retention node), a location that meets the observation angle requirements is identified on the sub-area boundary, centered on the key monitoring node, as the observation anchor point. The specific method is to establish a polar coordinate system with the coordinates of the key monitoring node as the origin. Each boundary point on the sub-area boundary is traversed and the angle between the line connecting the boundary point and the key monitoring node and the camera optical axis is calculated. If the angle is within the preset observation angle range and the boundary point is located on the sub-area boundary, it is selected as a candidate observation anchor point. Ultimately, the candidate observation anchor point with the least field of view obstruction and the widest coverage is selected as the official observation anchor point, ensuring that the key monitoring node is within the camera's clear monitoring range.

[0110] The generation of redundant anchor points is based on the preset redundant distance and spatial orientation information. The preset redundant distance is the backup anchor point spacing set to deal with camera failure or view obstruction, for example, it is set to 8-15 meters. On the boundary of the sub-area, with the key monitoring node as the benchmark, a point is selected along the boundary line at every preset redundant distance as a redundant anchor point. When selecting, it is necessary to avoid overlapping of redundant anchor points with observation anchor points, and ensure that the connection line with the key monitoring node meets the minimum monitoring distance requirement of the camera (such as not less than 2 meters). The role of the redundant anchor point is to automatically switch to the redundant anchor point to fill the blind spot when the observation anchor point cannot be effectively monitored due to equipment failure or environmental changes (such as tree obstruction), thereby improving the reliability of the monitoring system.

[0111] After generating a video surveillance path, the system uses real-time positioning technology (such as a video-based target tracking algorithm) to determine the real-time location of the individual and calculates the offset distance between each node and the individual's real-time location. This offset distance is calculated based on the Euclidean distance formula, which is the square root of the sum of the squares of the coordinate differences between two points. A preset tolerance threshold (for example, 5 meters) is set based on the required monitoring accuracy. When the offset distance exceeds this threshold, it indicates that the node location of the current surveillance path deviates significantly from the actual activity area of ​​the individual, triggering a path update mechanism.

[0112] The first step in path updating is to dynamically update the spatial coordinates of the observation anchor point. The system adjusts the position of the observation anchor point based on the relative position of the person's real-time location and the preset activity area. For example, if the person is concentrated near the northeastern boundary of the sub-area, and the original observation anchor point is located at the southwest boundary, resulting in an excessive offset distance, the observation anchor point will be moved toward the northeastern boundary until the offset distance from the person's real-time location is less than the fault tolerance threshold. During the movement process, the observation anchor point must be kept on the sub-area boundary, and its observation angle with key monitoring nodes must be recalculated to ensure that the monitoring field of view covers the key area.

[0113] After an observation anchor point is updated, the locations of redundant anchor points are recalculated based on the new observation anchor point location. The spacing and number of redundant anchor points are adjusted based on the updated observation anchor point distribution. For example, redundant anchor points are densely packed near the new observation anchor point to enhance surveillance coverage in that area. After the anchor point update is complete, the system iteratively optimizes the entire video surveillance path, rearranging the order of association between key monitoring nodes, observation anchor points, and redundant anchor points to ensure path continuity and surveillance efficiency.

[0114] To dynamically adjust the scope of pre-set activity areas, the system first uses IoT sensors and community management systems to acquire real-time data on building layout changes, such as new buildings, temporary barriers, and road closures. Based on the modified topology, a virtual space grid is generated, dividing the community space into regular or irregular grid cells (such as a 1m x 1m square grid), each of which corresponds to a unique coordinate identifier.

[0115] By analyzing the difference between historical and real-time data on the density distribution of people in a virtual space grid, changes in activity area usage can be determined. Historical data represents the average density of each grid cell over a period of time (e.g., a week), while real-time data represents the density value for the current period. The difference is calculated as the absolute difference between the real-time and historical densities. If the difference in a particular area exceeds a set density difference threshold (e.g., 1.5 people / square meter), it indicates a significant change in activity levels in that area. For example, if the historical density of a parking lot during off-peak hours is 0.2 people / square meter and the real-time density is 1.8 people / square meter, the difference is 1.6, exceeding the density difference threshold, indicating that the area may be temporarily used as a venue for activities and that the activity area needs to be expanded.

[0116] The boundary threshold of the activity zone is dynamically expanded or contracted based on the sign of the difference value. If the difference value is positive (real-time density is higher than historical density), the boundary of the area is expanded outward by a certain distance (for example, 3 meters); if it is negative (real-time density is lower than historical density), the boundary is contracted. After the boundary adjustment, it is superimposed with the preset minimum safety distance (for example, 2 meters) to generate the updated spatial demarcation parameters. The preset minimum safety distance ensures that the activity zone maintains a safe distance from community boundaries, buildings, and other areas to prevent monitoring from covering dangerous areas.

[0117] In practice, dynamic adjustments to activity zones need to be integrated with community functional zoning regulations. For example, the activity zone in a residential area automatically contracts at night, retaining only monitoring of main roads and public facilities; during the day, it expands to encompass all open spaces. The adjusted spatial demarcation parameters are automatically synchronized to the boundary calculation modules for each sub-area, triggering recalculation of the sub-area divisions and recollection of monitoring data.

[0118] The algorithms for anchor point generation and path updating are based on computer graphics and optimization theory. A greedy algorithm is used to select observation anchor points, selecting the local optimal solution among candidate observation anchor points. A uniform distribution strategy is used for the layout of redundant anchor points to ensure balanced boundary coverage. Path optimization uses an approximation algorithm for the Traveling Salesman Problem (TSP), such as the 2-opt heuristic algorithm, to reduce computational complexity while ensuring monitoring efficiency.

[0119] The system displays anchor point locations and monitoring paths through a visual interface, allowing monitoring personnel to intuitively view the distribution of anchor points and path directions in each sub-area and manually fine-tune anchor point locations to suit specific monitoring needs. For example, in areas shadowed by high-rise buildings, the observation anchor points can be manually adjusted to avoid fixed obstructions.

[0120] Example 5:

[0121] Regarding the interaction anomaly index correction and path backtracking mechanism, the specific implementation methods are as follows:

[0122] When calculating the interaction anomaly index, in addition to the duration difference, corrections must be made based on the data on the amplitude of the person's limb movements. First, the limb movement amplitude data of the participants in each interaction event cluster is extracted through a skeletal key point detection algorithm (such as OpenPose), including parameters such as joint angle change and limb movement speed. The preset action standard library stores feature models of various normal and abnormal actions, such as the range of joint angle changes for normal actions such as "waving" and "running", as well as feature thresholds for abnormal actions such as "pushing" and "climbing". The system matches the detected limb movement data with the models in the standard library, calculates the cosine similarity of the action feature vector, and if the similarity exceeds the preset matching threshold (such as 0.7), it is determined to be an abnormal action pattern.

[0123] For successfully matched abnormal action patterns, a weighted sum is calculated based on their risk level to generate an action risk coefficient. The risk level is divided into three levels: low risk (such as a slight push, weight 0.3), medium risk (such as a threat with a weapon, weight 0.6), and high risk (such as a weight 1.0). The action risk coefficient is calculated as follows:

[0124]

[0125] in, is the action risk coefficient, is the number of abnormal action modes, For the The danger level weight of each abnormal action, For the The matching similarity of abnormal actions.

[0126] The action risk coefficient and the duration difference are linearly fused to obtain the correction value of the interaction anomaly index, and the correction value is used to determine the secondary anomaly area. The linear fusion formula is:

[0127]

[0128] in, is the modified interaction anomaly index, is the duration difference (unit: minutes), is the fusion weight coefficient (value range 0~1, default value 0.5). Through this correction mechanism, the interaction anomaly index can more comprehensively reflect the risk of the behavior. For example, if the duration difference of an interaction event cluster is 3 minutes and it contains 2 medium-risk abnormal actions (similarity is 0.8 and 0.7 respectively), the action risk coefficient is , the modified interaction anomaly index .

[0129] During the dynamic path adjustment process, the system monitors the direction deviation change rate of high-risk behavior nodes in real time. The direction deviation change rate is the ratio of the difference in direction deviations in adjacent time intervals to the time interval, and the calculation formula is:

[0130]

[0131] in, is the rate of change of direction deviation (unit: degrees / second), is the direction deviation at the current moment, is the direction deviation at the previous moment, is the time interval (unit: seconds). The preset fluctuation threshold is an empirical parameter, for example, it is set to 5 degrees / second. When the preset fluctuation threshold is exceeded, it indicates that the movement direction of the high-risk behavior node has changed dramatically, and there may be external interference or escalation of abnormal behavior. At this time, the path backtracking mechanism is triggered.

[0132] After the path backtracking mechanism is activated, the system extracts scan records for clustered and lingering nodes from the previous period (e.g., the past 10 minutes), including data such as scan time, dwell time, and cluster risk coefficient for each node. Based on this historical data, the duration weights for the first scan sequence (clustered nodes) and the second scan sequence (lingering nodes) are recalculated. The duration weights are updated by dynamically adjusting the duration contribution of each node type based on the incidence of abnormal events in historical scans. For example, if a high proportion of clustered nodes trigger abnormal events in historical data, the duration weight of the first scan sequence is increased to 70%, while if a high proportion triggers abnormal events, the duration weight is reduced to 30%.

[0133] The recalculated duration weight is interpolated and fused with the duration weights of the current first scanning sequence and the second scanning sequence to generate a new weight for anti-interference. The scanning order and scanning duration are re-determined with the new weight for anti-interference. Interpolation fusion is achieved through a linear interpolation algorithm. For example, the recalculated duration weight is (first scan sequence) and (Second scanning sequence), the duration weights of the current first scanning sequence and the second scanning sequence are and , then the new weight is , ,in is the interpolation coefficient (range 0~1, default 0.5), and By integrating historical experience and current data, the scanning sequence is made more resistant to interference and avoids misjudgment of monitoring strategies caused by short-term data fluctuations.

[0134] At the system implementation level, the interaction anomaly index correction and path backtracking mechanism are implemented collaboratively through the real-time data processing module and the historical data query module. Skeletal key point detection and motion matching algorithms run on edge computing nodes to ensure real-time performance. Historical scan records are stored in a time-series database, enabling fast query and analysis. Monitoring personnel can review the interaction anomaly index correction process and path backtracking records through system logs, facilitating post-audit and policy optimization.

[0135] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0136] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

[0137] Any matters not described in the present invention are applicable to the prior art.

Claims

1. A method for video monitoring abnormal behavior of community personnel based on human factors engineering, characterized by: The method comprises: The community surveillance video data is spatially divided based on the preset activity area range to obtain the surveillance data of each sub-area, wherein the surveillance data of each sub-area includes personnel trajectory data and environmental interaction data; Determine the abnormality level of each sub-area based on the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data in each sub-area; Calculate the key monitoring nodes of each sub-area based on the anomaly level, personnel trajectory data, and environmental interaction data of each sub-area, and determine the scanning order and scanning duration of the key monitoring nodes based on dynamic path adjustment; Generate observation anchor points for each sub-area at the boundary of each sub-area based on the preset observation angle, key monitoring nodes of each sub-area, and spatial orientation information, and generate redundant anchor points for each sub-area at the boundary of each sub-area based on the preset redundant distance, the key monitoring nodes, and spatial orientation information; The key monitoring nodes, observation anchor points and redundant anchor points are associated and integrated based on the scanning order and scanning duration of the key monitoring nodes to generate a video monitoring path for the community.

2. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 1 is characterized in that: The community surveillance video data is spatially divided based on the preset activity area range to obtain the surveillance data of each sub-area, including: For the current sub-region, the division boundary of the current sub-region is calculated according to the geometric center coordinates of the current sub-region, the boundary coordinates of the adjacent sub-regions and the preset spatial overlap threshold; Extracting surveillance video data of consecutive frames based on the division boundary of the current sub-region and a preset time window; The movement trajectories of personnel and the environmental interaction events that conform to the division boundary in the monitoring video data of the continuous frames are marked as the monitoring data of the current sub-area.

3. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 1 is characterized in that: The abnormality level of each sub-area is determined based on the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data in each sub-area, including: For the current sub-region, when the number of movement per unit time of the personnel trajectory data exceeds a preset frequency threshold, the current sub-region is marked as a first-level abnormal region; When the number of movements per unit time does not exceed the preset frequency threshold, performing spatiotemporal clustering on the environmental interaction data based on a preset clustering radius to obtain interaction event clusters; The difference between the duration of each interaction event cluster and the preset benchmark duration is recorded as the duration difference, and the interaction anomaly index is calculated based on the duration difference; If the interaction anomaly index exceeds a preset anomaly threshold, marking the current sub-region as a secondary anomaly region; If the preset abnormal threshold is not exceeded, the current sub-area is marked as a regular monitoring area.

4. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 3 is characterized in that: The monitoring data of each sub-area also includes wear feature data and group gathering data. The key monitoring nodes include high-risk behavior nodes, retention nodes, and group gathering nodes. The key monitoring nodes of each sub-area are calculated based on the abnormality level, personnel trajectory data, and environmental interaction data of each sub-area, including: For the current sub-region, the group aggregation data is screened based on a preset density threshold to obtain a high-density aggregation region; Calculate the gathering risk coefficient based on the uniformity of personnel distribution and consistency of movement direction in the high-density gathering area; If the aggregation risk coefficient exceeds a preset risk threshold, the geometric center of the high-density aggregation area is determined as the group aggregation node; The coordinate points whose stay time exceeds a preset time length are extracted from the personnel trajectory data, and are marked as the stay nodes in combination with the interaction type of the environmental interaction data.

5. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 4 is characterized in that: After calculating the key monitoring nodes, the method further includes: generating a candidate set of high-risk behavior nodes based on the occurrence frequency and spatial distribution density of abnormal wearing identifiers in the wearing feature data; Performing redundancy filtering based on the spatiotemporal correlation of each candidate node in the candidate set of high-risk behavior nodes, and retaining candidate nodes that meet a preset correlation strength; The spatial coordinates of the retained candidate nodes that meet the preset association strength are vector-matched with the movement direction of the personnel trajectory data, and the candidate nodes with a direction deviation greater than a preset angle are screened out as the final high-risk behavior nodes.

6. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 5 is characterized in that: The dynamic path adjustment includes: Arranging the clustered nodes in descending order of clustering risk coefficients to generate a first scanning sequence; Arrange the stranded nodes in descending order according to the ratio of the stay time to the preset time length to generate a second scanning sequence; The total scanning time is set, and the two scanning sequences are allocated the total scanning time according to the time weight. The time weights of the first scanning sequence and the second scanning sequence are dynamically adjusted based on the average value of the difference between the direction deviation of all high-risk behavior nodes and the preset angle. The scanning time of the first scanning sequence and the second scanning sequence is updated, and the monitoring scan is performed with the updated scanning time of the two scanning sequences. The scanning sequence with a large time weight is scanned first to obtain the scanning order and scanning time of the key monitoring nodes.

7. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 6 is characterized in that: The dynamic path adjustment further includes: Real-time monitoring of the direction deviation change rate of the high-risk behavior node, and triggering a path backtracking mechanism when the direction deviation change rate exceeds a preset fluctuation threshold; Extracting the scanning records of the group gathering nodes and the stranded nodes in the previous period based on the path backtracking mechanism, and recalculating the duration weight; The recalculated duration weight is interpolated and fused with the duration weights of the current first scanning sequence and the second scanning sequence to generate a new anti-interference weight, and the scanning order and scanning duration are re-determined with the new anti-interference weight.

8. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 3 is characterized in that: The calculation of the interaction anomaly index further includes: Extracting the body movement amplitude data of the participants in each interaction event cluster, and matching abnormal movement patterns based on a preset movement standard library; The successfully matched abnormal action patterns are weighted and summed according to the danger level to generate the action risk coefficient; The action risk coefficient and the duration difference are linearly fused to output a corrected value of the interaction anomaly index.

9. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 1, characterized in that: After generating the video surveillance path, it also includes: Real-time detection of the offset distance between each key monitoring node in the video monitoring path and the real-time position of the personnel; If the offset distance exceeds a preset fault tolerance threshold, the spatial coordinates of the observation anchor point are dynamically updated according to the relative orientation between the real-time position of the person and the preset activity area; The generation position of the redundant anchor point is recalculated based on the updated observation anchor point, and the video monitoring path is iteratively optimized.

10. The method for video monitoring abnormal behavior of community personnel based on human factors engineering according to claim 1, characterized in that: The preset activity area range is dynamically adjusted, including: Obtain real-time data on building layout changes within the community and generate a virtual space grid based on the changed topological structure; Dynamically expand or contract the boundary threshold of the activity area according to the difference between the historical data and the real-time data of the density distribution of people in the virtual space grid; The boundary threshold is superimposed on the preset minimum safety distance to generate an updated space division parameter.

Images