Community personnel abnormal behavior video monitoring method based on human factor engineering

Abstract

The invention relates to the technical field of community video monitoring, and discloses a community personnel abnormal behavior video monitoring method based on human factor engineering, and the method comprises the steps: dividing community monitoring video data based on a preset activity region, and obtaining the monitoring data of each sub-region, including a personnel track, environment interaction and the like; determining the abnormal level of each sub-region according to the movement frequency of the personnel trajectory data and the interaction strength of the environment interaction data; key monitoring nodes are calculated in combination with abnormal levels and the like, and the scanning sequence and the scanning duration are adjusted and determined according to a dynamic path; then generating observation anchor points and redundant anchor points at the boundaries of the sub-regions; and finally, associating and integrating to generate a monitoring path. According to the method, through multi-dimensional data processing and dynamic optimization, the accuracy, flexibility and environmental adaptability of community abnormal behavior monitoring are improved, and the method is suitable for community safety management.

Description

Technical Field

[0001] The present invention relates to the technical field of community video surveillance, and particularly to a method for video surveillance of abnormal behaviors of community personnel based on ergonomics. Background Art

[0002] With the acceleration of the urbanization process, the composition of community personnel has become increasingly complex. The traditional video surveillance mode with fixed positions and single perspectives has gradually revealed many limitations and is difficult to meet the requirements of modern community security management for accurate identification and dynamic monitoring of abnormal behaviors of personnel.

[0003] In terms of monitoring scope and accuracy, traditional methods usually monitor the community as a whole, lacking refined partitioning of space. This results in mixed monitoring data and makes it difficult to quickly locate abnormal situations in specific areas. For example, in different scenarios such as crowded squares or long and narrow corridors, the behavior patterns of people are significantly different, but traditional monitoring cannot conduct targeted monitoring according to the characteristics of different areas, and it is easy to miss abnormal behaviors in local areas, such as small-scale group gatherings or individuals staying for a long time.

[0004] In terms of the identification of abnormal behaviors, most of the existing technologies rely on the extraction of behavior features with fixed thresholds and do not fully consider the interaction relationship between the environment and personnel behaviors in ergonomics. For example, only judge abnormalities by the moving speed or staying time of personnel, while ignoring the influence of environmental factors (such as building layout, facility distribution) on personnel behaviors. In the community, in some areas (such as monitoring blind spots or facility concentration points), due to environmental characteristics, people may have short-term abnormal interaction behaviors. Traditional methods are difficult to distinguish normal behaviors from real abnormal behaviors and are prone to false alarms or missed alarms.

[0005] In terms of the planning of monitoring paths, traditional methods mostly adopt scanning modes with fixed sequences and durations and lack a dynamic adjustment mechanism. When sudden abnormal events occur in the community, they cannot quickly optimize the monitoring path according to real-time data, resulting in a lag in monitoring key nodes. For example, when a group gathering suddenly appears in a certain area, the fixed scanning sequence and duration may not be able to include this area in key monitoring in time, delaying the discovery and handling of abnormal behaviors.

[0006] In addition, the existing technologies have insufficient adaptability to the dynamic changes of the community environment. Changes in the building layout within the community (such as adding temporary facilities or reconstructing roads) will change the regular paths and gathering areas of personnel activities, but traditional monitoring systems are difficult to update the monitoring scope and parameters in real time, resulting in loopholes in monitoring coverage. For example, a newly added express delivery stacking point may become a new personnel gathering area. If the traditional system does not adjust the monitoring strategy in time, it may not be able to effectively monitor the abnormal behaviors in this area.

[0007] In terms of data fusion and processing, traditional methods usually analyze personnel trajectory data or environmental data in isolation, without fully exploring the correlation information between the two. For example, a sudden change in a person's trajectory may be related to obstacles or abnormal events in the environment, but traditional methods cannot quickly identify such potential anomalies through correlation analysis, limiting the comprehensive judgment ability of complex abnormal behaviors. Summary of the Invention

[0008] The purpose of the present invention is to provide a video surveillance method for abnormal behaviors of community personnel based on ergonomics to solve the problems raised in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solutions: A video surveillance method for abnormal behaviors of community personnel based on ergonomics, the method includes: Spatially divide the community surveillance video data based on a preset activity area range to obtain the surveillance data of each sub-area, and the surveillance data of each sub-area includes personnel trajectory data and environmental interaction data; Determine the abnormal level of each sub-area according to the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data in each sub-area; According to the abnormal level, personnel trajectory data and environmental interaction data of each sub-area, calculate the key monitoring nodes 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 according to 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 according to the preset redundant distance, the key monitoring nodes and spatial orientation information; Integrate the key monitoring nodes, observation anchor points and redundant anchor points based on the scanning order and scanning duration of the key monitoring nodes to generate a video surveillance path for the community.

[0010] Further, spatially dividing the community surveillance video data based on a preset activity area range to obtain the surveillance data of each sub-area includes: For the current sub-area, calculate the division boundary of the current sub-area according to the geometric center coordinates of the current sub-area, the boundary coordinates of adjacent sub-areas and a preset spatial overlap threshold; Extract consecutive frames of surveillance video data based on the division boundary of the current sub-area and a preset time window; Mark the personnel movement trajectories and environmental interaction events that conform to the division boundary in the consecutive frames of surveillance video data as the surveillance data of the current sub-area.

[0011] Further, determine the anomaly level of each sub-region according to the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data, including: For the current sub-region, when the number of movements per unit time in the personnel trajectory data exceeds the preset frequency threshold, mark the current sub-region as a first-level anomaly region; When the number of movements per unit time does not exceed the preset frequency threshold, perform spatio-temporal clustering on the environmental interaction data based on a preset clustering radius to obtain each interaction event cluster; Record the difference between the duration of each interaction event cluster and the preset reference duration as the duration difference, and calculate the interaction anomaly index according to the duration difference; If the interaction anomaly index exceeds the preset anomaly threshold, mark the current sub-region as a second-level anomaly region; If it does not exceed the preset anomaly threshold, mark the current sub-region as a conventional monitoring region.

[0012] Further, the monitoring data of each sub-region further includes wearable feature data and group aggregation data, and the key monitoring nodes include high-risk behavior nodes, stay nodes, and group aggregation nodes. Calculate the key monitoring nodes of each sub-region according to the anomaly level, personnel trajectory data, and environmental interaction data of each sub-region, including: For the current sub-region, screen the group aggregation data based on a preset density threshold to obtain a high-density aggregation region; Calculate the aggregation risk coefficient according to the personnel distribution uniformity and movement direction consistency of the high-density aggregation region; If the aggregation risk coefficient exceeds the preset risk threshold, determine the geometric center of the high-density aggregation region as the group aggregation node; Extract the coordinate points in the personnel trajectory data where the stay time exceeds the preset duration, and mark them as the stay nodes in combination with the interaction type of the environmental interaction data.

[0013] Further, after calculating the key monitoring nodes, it further includes: Generate a candidate set of high-risk behavior nodes based on the occurrence frequency and spatial distribution density of abnormal wear identifiers in the wearable feature data; Perform redundancy filtering on each candidate node in the candidate set of high-risk behavior nodes according to the spatio-temporal correlation, and retain the candidate nodes that meet the preset correlation intensity; Perform vector matching on the spatial coordinates of the retained candidate nodes that meet the preset correlation intensity and the movement direction of the personnel trajectory data, and screen out the candidate nodes with a direction deviation greater than the preset angle as the final high-risk behavior nodes.

[0014] Further, the dynamic path adjustment includes: Arrange the group gathering nodes in descending order of the aggregation risk coefficient to generate a first scanning sequence; Arrange the staying nodes in descending order of the ratio of the staying time to the preset duration to generate a second scanning sequence; Set the total scanning duration. The two scanning sequences allocate the total scanning duration according to the duration weight. Dynamically adjust the duration weights of the first scanning sequence and the second scanning sequence based on the average value of the difference between the direction deviation degree of all high-risk behavior nodes and the preset angle, and update the scanning durations of the first scanning sequence and the second scanning sequence. Perform monitoring scans with the updated scanning durations of the two scanning sequences, and preferentially scan the scanning sequence with a larger duration weight to obtain the scanning order and scanning duration of the key monitoring nodes.

[0015] Furthermore, the dynamic path adjustment further includes: Real-time monitor the change rate of the direction deviation degree of the high-risk behavior nodes. When the change rate of the direction deviation degree exceeds the preset fluctuation threshold, trigger the path backtracking mechanism; Extract the scanning records of the group gathering nodes and staying nodes in the previous period based on the path backtracking mechanism, and recalculate the duration weights; Interpolate and fuse the recalculated duration weights with the current duration weights of the first scanning sequence and the second scanning sequence to generate anti-interference new weights, and re-determine the scanning order and scanning duration with the anti-interference new weights.

[0016] Furthermore, the calculation of the interaction anomaly index further includes: Extract the limb movement amplitude data of the participants in each interaction event cluster, and match the abnormal movement patterns based on the preset movement standard library; Sum up the matched abnormal movement patterns by weight according to the danger level to generate an action risk coefficient; Linearly fuse the action risk coefficient with the duration difference to output the corrected value of the interaction anomaly index.

[0017] Furthermore, after generating the video monitoring path, it further includes: Real-time detect 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 the preset fault tolerance threshold, dynamically update the spatial coordinates of the observation anchor point according to the relative orientation between the real-time position of the personnel and the preset activity area range; Recalculate the generation position of the redundant anchor point based on the updated observation anchor point, and iteratively optimize the video monitoring path.

[0018] Furthermore, the dynamic adjustment of the preset activity area range includes: Obtain the building layout change data in the community in real time, and generate a virtual space grid based on the changed topological structure; According to the difference value between the historical data and the real-time data of the personnel density distribution in the virtual space grid, dynamically expand or contract the boundary threshold of the activity area range; Overlay and calculate the boundary threshold with the preset minimum safety distance to generate updated space division parameters.

[0019] Compared with the prior art, the beneficial effects of the present invention are: In terms of space division and data processing, the community surveillance video data is spatially divided through a preset activity area range, and the sub-region division boundary is determined by combining the geometric center coordinates, adjacent boundary coordinates and the preset space overlap threshold to ensure the independence and relevance of the surveillance data of each sub-region. Extract continuous frame surveillance video data based on the division boundary and time window, and mark the personnel movement trajectories and environmental interaction events that meet the boundary, realizing the refined monitoring of the community space, avoiding the problem of data mixing in the traditional overall monitoring mode, and providing an accurate data basis for subsequent abnormal behavior analysis.

[0020] The abnormal level assessment combines the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data. By setting the frequency threshold and spatio-temporal clustering analysis, it can accurately distinguish the first-level abnormal area, the second-level abnormal area and the conventional monitoring area. At the same time, the wearable feature data and the group aggregation data are introduced to further refine the calculation of the key monitoring nodes (high-risk behavior nodes, staying nodes, group aggregation nodes). For example, the high-density aggregation area is screened by presetting the density threshold, and the aggregation risk coefficient is calculated by combining the personnel distribution uniformity and the movement direction consistency, ensuring the comprehensive monitoring of different types of abnormal behaviors, improving the accuracy and comprehensiveness of abnormal behavior recognition, and reducing false alarms and missed alarms.

[0021] The scanning order and scanning duration of the key monitoring nodes are generated based on dynamic path adjustment. The group aggregation nodes are arranged in descending order of the aggregation risk coefficient, the staying nodes are arranged in descending order of the ratio of the staying time to the preset duration, and the duration weight is dynamically adjusted according to the direction deviation degree of the high-risk behavior nodes. 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 the real-time abnormal situation, 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 change rate of the direction deviation degree of the high-risk behavior node exceeds the preset fluctuation threshold, the duration weight is recalculated and a new anti-interference weight is generated, enhancing the stability and anti-interference ability of the monitoring path.

[0022] The generation mechanism of the observation anchor points and redundant anchor points combines the preset observation angles, redundant distances, and spatial orientation information to generate anchor points at the boundaries of sub-regions and associate and integrate them into the video surveillance path. At the same time, by real-time detecting the offset distance between the key surveillance nodes and the real-time positions of personnel, the positions of the observation anchor points and redundant anchor points are dynamically updated, and the surveillance path is iteratively optimized, enabling the surveillance system to adapt to the changes in personnel positions in real time, ensuring the accuracy and continuity of the surveillance footage, and avoiding surveillance blind spots caused by personnel movement.

[0023] The dynamic adjustment function of the preset activity area range is based on the community building layout change data and personnel density distribution data. By generating virtual space grids and adjusting the boundary thresholds, it realizes the self-adaptation to the dynamic changes in the community environment. This dynamic adjustment ensures that the surveillance range always covers the actual activity areas of personnel, avoids surveillance loopholes caused by environmental changes, and improves the environmental adaptability and long-term effectiveness of the surveillance system.

[0024] The corrected calculation of the interaction anomaly index introduces the limb movement amplitude data and the preset action standard library. By matching the abnormal action patterns and weighted calculating the action risk coefficient, and linearly fusing with the duration difference to generate a correction value, it further improves the recognition accuracy of potential abnormal behaviors in the environmental interaction data, enables the system to more accurately judge the danger of interaction behaviors, and provides a more reliable basis for the early warning of abnormal behaviors. Description of the Drawings

[0025] Figure 1 It is a schematic flow chart of the method for video surveillance of abnormal behaviors of community personnel based on human factors engineering of the present invention; Figure 2 It is a schematic diagram of the principle of the spatial division of community surveillance video data; Figure 3 It is a schematic diagram of the principle calculation of key surveillance nodes (crowd gathering nodes, staying nodes); Figure 4 It is a schematic flow chart of dynamic path adjustment. Detailed Embodiment

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] Please refer to Figures 1-4 , a method for video surveillance of abnormal behaviors of community personnel based on human factors engineering involved in the present invention, the specific implementation steps are as follows: Spatially divide the community surveillance video data based on a preset activity area range to obtain the surveillance data of each sub-region, where the surveillance data includes personnel trajectory data and environmental interaction data. Through the structured segmentation of the community physical space, refined surveillance of personnel activities in different regions is achieved.

[0028] Determine the anomaly level of each sub-region according to the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data in each sub-region. By quantitatively analyzing the personnel movement characteristics and environmental interaction patterns, identify the areas that need to be focused on.

[0029] Calculate the key surveillance nodes of each sub-region based on the anomaly level, personnel trajectory data, and environmental interaction data of each sub-region, and determine the scanning order and scanning duration of the key surveillance nodes based on dynamic path adjustment. By establishing a multi-level key surveillance node system, optimize the allocation of surveillance resources.

[0030] Generate observation anchor points at the boundaries of each sub-region according to the preset observation angle, key surveillance nodes of each sub-region, and spatial orientation information; at the same time, generate redundant anchor points at the boundaries of each sub-region according to the preset redundant distance, key surveillance nodes, and spatial orientation information. By setting anchor points with different functions, improve the comprehensiveness and reliability of surveillance coverage.

[0031] Integrate and associate the key surveillance nodes, observation anchor points, and redundant anchor points based on the scanning order and scanning duration of the key surveillance nodes to generate the video surveillance path of the community. By constructing a dynamically adjustable video surveillance path, achieve efficient capture and tracking of abnormal behaviors.

[0032] Embodiment 1: Based on the above overall solution, the specific implementation of the spatial division is as follows: For the current sub-region, it is first necessary to determine its geometric center coordinates, which can be obtained by locating in the spatial coordinate system of the community. For example, based on the Gaussian plane coordinate system or the UTM coordinate system, the sub-region is abstracted into a geometric shape (such as a rectangle, a circle, etc.) and the coordinates of its geometric center point are calculated. The boundary coordinates of adjacent sub-regions are determined by the vertex coordinates of the geometric shape boundaries of other sub-regions that have been divided. The geometric shapes of each sub-region can be divided into irregular polygons according to the actual layout of the community (such as road and building distribution) to fit the real monitoring scenario. The preset spatial overlap threshold is a configurable distance parameter, for example, set to 1 - 5 meters, which is used to ensure that there is an overlapping area of monitoring coverage between adjacent sub-regions, and to avoid monitoring blind spots caused by camera viewing angle limitations or installation position deviations. When specifically calculating the division boundary, taking the geometric center coordinates of the current sub-region as the reference, combined with the boundary coordinates of adjacent sub-regions and the preset spatial overlap threshold, the boundary vertex coordinates of the current sub-region are determined through geometric operations (such as vector calculation, polygon intersection, etc.) to form a closed polygon division boundary. For example, if the distance from the boundary of an adjacent sub-region to the geometric center of the current sub-region is D meters and the preset overlap threshold is S meters, the boundary of the current sub-region can be extended S meters in the adjacent direction, so that the two sub-regions overlap within a range of S meters at the junction.

[0033] After the boundary division is completed, it is necessary to extract the monitoring video data of consecutive frames based on the division boundary of the current sub-region and the preset time window. The preset time window refers to the consecutive time period for analysis, which can be set to 5 minutes, 10 minutes, etc. according to the monitoring requirements. The system cuts the video stream through timestamps and extracts all video frames within this time period. During the extraction process, video analysis technology is used to process each frame of the consecutive frames. Through object detection algorithms (such as YOLO, SSD, etc.), the human targets in the frames are identified, and through object tracking algorithms (such as DeepSORT), the human movement trajectories are generated. For each trajectory, the system real-time determines whether the trajectory points are located within the division boundary of the current sub-region, which is specifically implemented through the inclusion detection algorithm of points and polygons (such as the ray method): draw a ray from the trajectory point in any direction and calculate the number of intersection points between the ray and the boundary polygon of the sub-region. If the number of intersection points is odd, it is determined that the point is inside the polygon, otherwise it is outside. The human movement trajectory data that conforms to the division boundary will be recorded, including parameters such as the coordinates, timestamps, and moving speeds of the trajectory points.

[0034] Meanwhile, the system detects and marks environmental interaction events in consecutive frames. Environmental interaction events include the interaction behaviors of people with objects or facilities in the community environment, such as pushing the access control, picking up items on the ground, approaching dangerous areas (such as pools, electrical boxes), etc. Through image recognition technology and a preset interaction behavior rule library, the system matches and identifies the actions of people and environmental elements in the video frames. For example, when it detects that a person's hand approaches an object on the ground and a grasping action occurs, it is marked as an "item pickup" event; when a person continuously approaches the boundary of a dangerous area equipped with sensors, it is marked as a "dangerous area approach" event. The occurrence time, location, and type of these interaction events will be recorded as environmental interaction data.

[0035] In practical applications, the division of sub - regions can be set differently according to the functional areas of the community. For example, for the main road area with dense personnel flow, it can be divided into smaller sub - regions (such as polygons with a side length of 10 meters) to improve the monitoring accuracy; for relatively open areas such as green belts and parking lots, it can be divided into larger sub - regions (such as polygons with a side length of 20 meters) to reduce the consumption of computing resources. When calculating the division boundaries, the installation position and viewing range of the camera need to be considered to ensure that each sub - region is covered by the effective monitoring range of at least one camera. For example, if a camera has a horizontal viewing angle of 120 degrees, a vertical viewing angle of 60 degrees, and an effective monitoring distance of 30 meters, then with this camera as the center, the division boundary of the sub - region within its viewing range should be within the viewing coverage area at a distance of 30 meters.

[0036] In addition, the system needs to regularly calibrate the division boundaries to cope with the changes in the monitoring range caused by community environmental changes (such as temporary obstacle settings, road construction). During the calibration process, the boundary coordinates of adjacent sub - regions are updated manually or automatically, and the division boundary of the current sub - region is recalculated to ensure the accuracy of the spatial division. When extracting monitoring data, for the trajectory of a person moving across sub - regions, the system will automatically segment the trajectory data according to the time sequence, and assign the trajectory points within the boundary of the current sub - region during different time periods to the personnel trajectory data of the current sub - region, while the trajectory points outside the boundary are assigned to the data of adjacent sub - regions, realizing the seamless connection and accurate classification of cross - region trajectories.

[0037] The marking of environmental interaction data needs to be filtered in combination with context information to avoid false alarms. For example, for the "item pickup" event, if it is detected that the item picked up by a person is a common daily necessity (such as a schoolbag, a water bottle) and the person leaves normally after picking it up, it may be determined as a normal behavior; if the item picked up is a suspicious package and the person stays for a long time or moves to a hidden area after picking it up, it is marked as a high - risk interaction event. The system classifies the risk levels of different types of interaction events by establishing a behavior pattern knowledge base, providing multi - dimensional data support for the determination of subsequent anomaly levels.

[0038] In terms of data storage, the monitoring data of each sub-region is stored in a time series format. The personnel trajectory data is indexed by trajectory ID, and each trajectory contains a series of coordinate points and corresponding timestamps. The environmental interaction data is indexed by event ID, and each event record contains the event type, occurrence time, location coordinates, and relevant video frame index. This storage structure facilitates subsequent data retrieval, analysis, and visualization. For example, the movement path of personnel within the sub-region can be reproduced through the trajectory playback function, and the occurrence frequency and spatial distribution of specific types of interaction events can be analyzed through the event statistics function.

[0039] Embodiment 2: Regarding the method for determining the anomaly level, the specific implementation process is as follows: For the current sub-region, first, a quantitative analysis is performed on the movement frequency in the personnel trajectory data. The personnel trajectory data contains the number of movements within a unit time, which is calculated based on 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 personnel movement frequency in the normal activity scenario of the community. For example, it is set to 5 times / minute (indicating a movement once every 12 seconds on average). When the number of movements per unit time in the personnel trajectory data of the current sub-region exceeds this preset frequency threshold, it indicates that the personnel movement is abnormally frequent, and there may be abnormal behaviors such as chasing or fleeing. At this time, the current sub-region is marked as a first-level anomaly region, triggering the highest-level monitoring response, such as increasing the monitoring scan frequency of this region, starting multi-camera linkage tracking, etc.

[0040] If the number of movements per unit time does not exceed the preset frequency threshold, it enters the spatio-temporal clustering analysis process of the environmental interaction data. The spatio-temporal clustering analysis processes the event points in the environmental interaction data based on a preset clustering radius. The preset clustering 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), which 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 their occurrences is less than or equal to the preset time interval, then they are considered to belong to the same interaction event cluster. Through density clustering algorithms such as DBSCAN, all event points in the environmental interaction data are clustered to obtain several interaction event clusters, and each cluster represents a set of interaction behaviors that occur within a similar spatio-temporal range.

[0041] Denote the difference between the duration of each interaction event cluster and the preset reference duration as the duration difference, and calculate the interaction anomaly index based on the duration difference. Calculate the difference between the duration of each interaction event cluster and the preset reference duration. The preset reference duration is a reference value set according to the average duration of normal interaction behaviors, for example, set to 5 minutes. For each interaction event cluster, its duration is the time difference between the last event and the first event in the cluster. If the duration of a certain cluster is T minutes, then the duration difference is |T - 5| minutes. This duration difference reflects the degree to which the duration of the interaction behavior in this cluster deviates from the normal level. The larger the difference, the more abnormal the interaction behavior. Calculate the average value of the duration differences of all interaction event clusters to obtain the interaction anomaly index. The preset anomaly threshold is a critical value obtained based on historical data statistics, for example, set to 2 minutes. When the average value of the duration differences of all interaction event clusters (i.e., the interaction anomaly index) exceeds the preset anomaly threshold, it indicates that there is an overall anomaly in the environmental interaction behavior of the current sub-region, and it is marked as a secondary anomaly area, and the monitoring attention needs to be increased; if it does not exceed, it is marked as a regular monitoring area and scanned at the standard frequency.

[0042] During the analysis process, the wearable feature data and crowd aggregation data in the monitoring data provide supplementary basis for the judgment of the anomaly level. The wearable feature data is extracted through image recognition technology, including the clothing color, style of the personnel, and whether they wear special identifiers (such as safety helmets, reflective vests), etc. For example, detecting that a person wears a safety helmet in a non-construction area, or detecting that a person wears emergency rescue clothing during normal periods can be regarded as an abnormal wearable identifier. The system records the occurrence times and location distributions of the abnormal wearable identifiers as auxiliary judgment factors. The crowd aggregation data is generated through a density detection algorithm. This algorithm is based on the personnel position distribution in the video frame and calculates the number of personnel per unit area. When the personnel density exceeds the preset density threshold (such as 3 people per square meter), it is determined as a high-density aggregation area.

[0043] For the determination of the first-level anomaly area, in addition to the movement frequency, the wearable feature data can also be combined for cross-verification. For example, if the movement frequency of the personnel in a certain sub-region exceeds the preset frequency threshold and multiple abnormal wearable identifiers are detected at the same time, it is further confirmed that there are high-risk abnormal behaviors in this area, improving the credibility of the anomaly level. For the secondary anomaly area, if multiple interaction behaviors involving abnormal wearable personnel are found in the interaction event cluster, then adjust the calculation weights of each interaction event cluster when calculating the interaction anomaly index, increasing the influence of this type of event on the anomaly level.

[0044] During the spatio-temporal clustering process, it is necessary to handle interaction events across sub-regions. If the event points of an interaction event cluster are distributed in multiple sub-regions, the system automatically identifies all the sub-regions involved in the cluster and separately includes the relevant data of the cluster in the calculation of the anomaly levels of each sub-region. For example, for a clustering event cluster that spans two sub-regions, the duration difference and the number of abnormal wearing identifiers will be respectively reflected in the calculation of the interaction anomaly index of the two sub-regions, avoiding the fragmentation of events caused by regional division.

[0045] In addition, the system supports dynamic adjustment of parameters such as the preset frequency threshold, the preset clustering radius, and the preset reference duration. For example, during the peak period of community activities (such as the morning and evening commuting hours), the normal movement frequency of people is relatively high, and the preset frequency threshold can be automatically increased; during the night time period, the activities of people decrease, and the preset frequency threshold can be decreased to improve the sensitivity to abnormal behaviors. The dynamic parameter adjustment rules are generated based on the historical data statistical model. By analyzing the behavior patterns of people in different time periods and different weather conditions, the parameter values are automatically optimized to enhance the adaptability of the anomaly level judgment.

[0046] In terms of data processing efficiency, for large-scale surveillance video data, a distributed computing framework is adopted to process the personnel trajectory data and the environmental interaction data in parallel. For example, the video data of different sub-regions are allocated to different computing nodes for trajectory extraction and event detection, and data synchronization and clustering result summary among the computing nodes are achieved through a message queue to ensure the rapid determination of the anomaly level under the real-time requirements.

[0047] The marked results of the anomaly levels are displayed in real time through a visualization interface. The community monitoring center can view the anomaly level status of each sub-region (for example, the first-level anomaly area is marked in red, the second-level in yellow, and the normal area in green), and automatically trigger the corresponding monitoring strategies according to the levels. For example, the first-level anomaly area triggers an audible and visual alarm and automatically saves the recent video recordings of this area; the second-level anomaly area starts slow-motion playback and key frame marking to facilitate the monitoring personnel to quickly locate abnormal events.

[0048] Example 3: Regarding the calculation of key monitoring nodes and dynamic path adjustment, the specific implementation methods are as follows: For the current sub-region, the determination of crowd gathering nodes requires two stages: data screening and risk assessment. First, the crowd gathering data is screened based on a preset density threshold. The preset density threshold is set according to the statistical value of the personnel density in the community's daily activities. For example, it is set to 2.5 people per square meter. When the personnel density in a certain area within the sub-region exceeds this preset density threshold, it is determined as a high-density gathering area. The personnel density is calculated based on the personnel detection results in the video frames. The specific method is as follows: The sub-region is divided into several grid units, each with an area of S2 square meters. The number of personnel N2 in each unit is counted, and the personnel density of this unit is N2 / S2 (people per square meter). By traversing all grid units, the areas with density exceeding the preset density threshold are screened out as high-density gathering areas.

[0049] Calculate the gathering risk coefficient of the high-density gathering area. This coefficient is comprehensively obtained through two indicators: personnel distribution uniformity and movement direction consistency. Personnel distribution uniformity reflects the degree of dispersion of the spatial distribution of the gathered crowd. The calculation formula is: the ratio of the standard deviation of the personnel density of each grid unit in the area to the average density. The larger the ratio, the more uneven the distribution and the higher the potential risk. Movement direction consistency is measured by calculating the average cosine similarity of the movement direction vectors of each person in the gathered crowd. The higher the similarity, the more consistent the movement direction of the crowd (such as moving in a collective direction towards a certain direction), which may indicate an abnormal event (such as a panic evacuation). The gathering risk coefficient combines the above two indicators through linear weighting. The weights can be set according to the characteristics of risk events in historical data. For example, the weight of personnel distribution uniformity accounts for 40%, and the weight of movement direction consistency accounts for 60%. If the gathering risk coefficient exceeds the preset risk threshold (such as 0.6), the geometric center of the high-density gathering area is determined as the crowd gathering node, and this crowd gathering node is included in the monitoring path as a key monitoring point.

[0050] The extraction of the staying nodes is based on the staying time in the personnel trajectory data. The system traverses all personnel trajectory points. For each trajectory point, the time interval from this point to the next trajectory point is calculated. If the time interval exceeds the preset duration (such as 30 minutes), then this trajectory point is determined as a staying point. The preset duration can be set according to the staying habits in the normal activities of the community. For example, the normal staying duration allowed in the square rest area is 20 minutes, so the preset duration is set to 30 minutes to distinguish abnormal stays. After extracting all the coordinate points with staying time exceeding the preset duration, they are marked in combination with the interaction types of the environmental interaction data: If there are frequent environmental interaction events near the staying point (such as operating public facilities multiple times), it may belong to a normal stay; if the staying point is located in a remote area and there is no obvious interaction behavior, it is marked as a staying node, indicating possible abnormal wandering or potential safety hazards.

[0051] 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 occurrence frequency and spatial distribution density of abnormal wearing identifiers in wearable feature data. Abnormal wearing identifiers include, but are not limited to, wearing work clothes during non-working hours, clothing that covers the face, carrying suspicious items, etc. The system uses an image recognition model (such as FasterR-CNN) to detect the wearing features of people in video frames. When the number of times an abnormal wearing identifier appears in the same area within a unit time exceeds a set threshold (such as 5 times per hour), or the spatial distribution of the abnormal wearing identifier shows aggregation (such as 3 cases appearing within a range of 10 square meters), the coordinate points of this area are added to the candidate set.

[0052] In the redundancy filtering stage, duplicate or irrelevant nodes are removed by analyzing the spatio-temporal correlation of each candidate node in the candidate set of high-risk behavior nodes. The spatio-temporal correlation calculation includes double constraints of time interval and spatial distance: 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 as associated nodes of the same high-risk behavior event, and only one candidate node (such as the earliest appearing node) is retained as a representative. Through this step, redundant nodes generated by the repeated detection of the same abnormal behavior are reduced, and the effectiveness of the node set is improved.

[0053] Finally, the spatial coordinates of the candidate nodes that meet the preset association strength are vector-matched with the moving direction of the personnel trajectory data. Specifically, for each candidate node, the moving direction vector of the personnel at its location (calculated from the front and back coordinates of the trajectory points) is obtained, and the angle between this vector and the direction vector from the candidate node to the geometric center of the sub-region is calculated as the direction deviation. If the angle is greater than the preset angle (such as 60 degrees), it indicates that the moving direction of the personnel significantly deviates from the normal activity direction of the sub-region, and there may be a deliberate avoidance of monitoring or an abnormal movement path. Such nodes are screened as the final high-risk behavior nodes.

[0054] Dynamic path adjustment realizes the optimal allocation of monitoring resources through multi-sequence priority management. First, the group aggregation nodes are sorted in descending order of the aggregation risk coefficient to generate the first scanning sequence, ensuring that high-risk aggregation areas are monitored first. For example, a node with an aggregation risk coefficient of 0.8 is ranked before a node with a coefficient of 0.7. The monitoring system scans each group aggregation node in this order to ensure that high-risk areas are covered in the shortest time.

[0055] The stay nodes are sorted in descending order of the ratio of the stay time to the preset duration to generate the second scanning sequence. The smaller this ratio, the shorter the stay time (for example, if the stay time is 40 minutes and the preset duration is 30 minutes, the ratio is 1.33; if the stay time is 50 minutes, the ratio is 1.67). After sorting in descending order, the stay nodes with longer stay times will be scanned first to promptly detect potential risks of long-term stays.

[0056] Set the total scanning duration. The two scanning sequences allocate the total scanning duration according to the duration weights. Dynamically adjust the duration weights of the first scanning sequence and the second scanning sequence based on the average value of the difference between the direction deviation degrees of all high-risk behavior nodes and the preset angle. Update the scanning durations of the first scanning sequence and the second scanning sequence, and perform monitoring scans with the updated scanning durations of the two scanning sequences. Prioritize the scanning of the scanning sequence with a larger duration weight to obtain the scanning order and scanning duration of the key monitoring nodes.

[0057] The direction deviation degree of the high-risk behavior nodes is used to dynamically adjust the duration weights of the first two scanning sequences. The specific method is as follows: Calculate the average value of the difference between the direction deviation degrees 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 behaviors with abnormal directions in the current sub-region, and it is necessary to increase the duration weight of the first scanning sequence (crowd aggregation nodes) (such as increasing from the default 50% to 60%) to cope with possible crowd anomalies. If the average value is less than or equal to 10 degrees, then maintain or reduce the duration weight of the first scanning sequence and increase the duration weight of the second scanning sequence to pay attention to the risk of individual stay. By adjusting the duration weights in real time, dynamic response to different types of anomalies is achieved.

[0058] At the system implementation level, the calculation of the key monitoring nodes, the generation of the scanning order and the scanning duration are completed by the real-time data processing module. This module receives the monitoring data from each sub-region, calculates parameters such as the aggregation risk coefficient and the direction deviation degree through multi-threaded parallel computing, and manages the scanning order of various types of key monitoring nodes using the priority queue data structure. After the monitoring path is generated, it is transmitted to the camera control system through the network to drive the camera to scan each key monitoring node in sequence, and at the same time record the scanning time and video data to form a complete monitoring log.

[0059] In addition, the system supports manual intervention in the scanning order. For example, when the monitoring personnel find a sudden anomaly in a certain area through the visualization interface, they can manually adjust the priority of the nodes in that area to force them to be scanned in advance. The manual intervention operation is realized through the event trigger mechanism to ensure a quick response in case of emergency.

[0060] Example 4: Regarding the generation of anchor points, the update of the monitoring path, and the dynamic adjustment of the activity area, the specific implementation is as follows: When generating observation anchor points, it is necessary to comprehensively consider the preset observation angle, the positions of key monitoring nodes, and the spatial orientation information. The preset observation angle is a physical parameter of the camera or a software-configurable parameter. For example, the optimal observation angle of a certain model of camera is 110 degrees horizontally and 70 degrees vertically. The system determines the fan-shaped area covered by the camera based on this angle. For each key monitoring node (such as a crowd gathering node, a staying node), taking the key monitoring node as the center, find a position on the boundary of the sub-region that meets the observation angle requirement as the observation anchor point. The specific method is as follows: taking the coordinates of the key monitoring node as the origin, establish a polar coordinate system, traverse each boundary point on the boundary of the sub-region, calculate the angle between the line connecting the boundary point and the key monitoring node and the optical axis of the camera. If the angle is within the preset observation angle range and the boundary point is on the boundary of the sub-region, then it is used as a candidate observation anchor point. Finally, select the candidate observation anchor point with the least field of view occlusion and the widest coverage range as the formal observation anchor point to ensure that the key monitoring node is within the clear monitoring range of the camera.

[0061] The generation of redundant anchor points is based on the preset redundant distance and spatial orientation information. The preset redundant distance is the spacing of backup anchor points set to cope with camera failures or perspective occlusions, for example, set to 8 - 15 meters. On the boundary of the sub-region, taking the key monitoring node as the reference, select a point as a redundant anchor point every preset redundant distance along the boundary line. When selecting, it is necessary to avoid the overlap of redundant anchor points and observation anchor points, and ensure that the line connecting it and the key monitoring node meets the minimum monitoring distance requirement of the camera (such as not less than 2 meters). The role of redundant anchor points is to automatically switch to redundant anchor points for blind spot filling when the observation anchor points cannot effectively monitor due to equipment failures or environmental changes (such as tree occlusion), improving the reliability of the monitoring system.

[0062] After generating the video monitoring path, the system obtains the real-time position of personnel through real-time positioning technology (such as video-based target tracking algorithms), and calculates the offset distance between each node and the real-time position of personnel. The calculation of the offset distance is based on the Euclidean distance formula, that is, the square root of the sum of the squares of the coordinate differences between two points. The preset fault tolerance threshold is set according to the monitoring accuracy requirements, for example, 5 meters. When the offset distance exceeds this preset fault tolerance threshold, it indicates that there is a significant deviation between the node position of the current monitoring path and the actual personnel activity area, and the path update mechanism needs to be triggered.

[0063] The first step in path update is to dynamically update the spatial coordinates of the observation anchor points. The system adjusts the position of the observation anchor points according to the relative orientation of the real-time position of personnel and the preset activity area range. For example, if the personnel are concentrated near the northeast boundary of the sub-region, while the original observation anchor point is located on the southwest boundary, resulting in an excessive offset distance, then move the observation anchor point towards the northeast boundary direction until the offset distance between it and the real-time position of personnel is less than the fault tolerance threshold. During the movement process, it is necessary to keep the observation anchor point on the boundary of the sub-region and recalculate its observation angle with the key monitoring node to ensure that the monitoring field of view covers the key area.

[0064] After the observation anchor points are updated, the generation positions of redundant anchor points are recalculated based on the new positions of the observation anchor points. The spacing and quantity of the redundant anchor points are adjusted according to the distribution of the updated observation anchor points. For example, redundant anchor points are encrypted near the new observation anchor points to enhance the monitoring coverage density in this area. After completing the anchor point update, the system iteratively optimizes the entire video monitoring path, rearranging the association order of key monitoring nodes, observation anchor points, and redundant anchor points to ensure the continuity of the path and the monitoring efficiency.

[0065] For the dynamic adjustment of the preset activity area range, the system first obtains the building layout change data in real time through interfaces such as Internet of Things sensors and community management systems, such as information on newly built buildings, temporary isolation fence settings, road closures, etc. Based on the changed topological structure, a virtual space grid is generated, that is, the community space is divided into regular or irregular grid cells (such as 1-meter by 1-meter square grids), and each grid cell corresponds to a unique coordinate identifier.

[0066] By analyzing the difference value between the historical data and real-time data of the personnel density distribution in the virtual space grid, the change in the use of the activity area is judged. The historical data is the average personnel density of each grid cell over a past period (such as 1 week), and the real-time data is the density value at the current time period. The difference value is calculated as the absolute difference between the real-time density and the historical density. If the difference value of a certain area exceeds the set density difference threshold (such as 1.5 people per square meter), it indicates that the personnel activity volume in this area has changed significantly. For example, the historical density of a parking lot area during off-peak hours is 0.2 people per square meter, the real-time density is 1.8 people per square meter, and the difference value is 1.6, exceeding the density difference threshold, indicating that this area may be temporarily used as an activity venue and the activity area range needs to be expanded.

[0067] According to the positive or negative of the difference value, the boundary threshold of the activity area range is dynamically expanded or contracted. If the difference value is positive (real-time density is higher than historical density), the boundary of this area is expanded outward by a certain distance (such as 3 meters); if it is negative (real-time density is lower than historical density), the boundary is contracted. After the boundary is adjusted, it is superimposed with the preset minimum safety distance (such as 2 meters) to generate the updated space division parameters. The preset minimum safety distance is used to ensure a safe distance between the activity area and the community boundary, buildings, etc., to avoid monitoring dangerous areas.

[0068] In practical applications, the dynamic adjustment of the activity area range needs to be combined with the functional zoning rules of the community. For example, the activity area range in the residential area automatically shrinks at night, only retaining the monitoring of the main roads and public facilities areas; during the day, it expands to the entire open space. The adjusted space division parameters are automatically synchronized to the boundary calculation modules of each sub-area, triggering the recalculation of the sub-area division and the re-acquisition of monitoring data.

[0069] The algorithm implementation of anchor point generation and path update is based on computer graphics and optimization theory. The selection of observation anchor points adopts a greedy algorithm to select the local optimal solution among candidate observation anchor points; the layout of redundant anchor points adopts a uniform distribution strategy to ensure the balance of boundary coverage; the path optimization adopts an approximation algorithm for the Traveling Salesman Problem (TSP), such as the 2-opt heuristic algorithm, to reduce the computational complexity while ensuring the monitoring efficiency.

[0070] The system displays the anchor point positions and monitoring paths through a visualization interface. Monitoring personnel can intuitively view the anchor point distribution and path directions in each sub-region and manually fine-tune the anchor point positions to meet special monitoring requirements. For example, in areas with shadows of high-rise buildings, manually adjust the positions of observation anchor points to avoid fixed obstacles.

[0071] Example 5: Regarding the correction of the interaction anomaly index and the path backtracking mechanism, the specific implementation method is as follows: When calculating the interaction anomaly index, in addition to the duration difference, it is necessary to combine the data of the amplitude of personnel's body movements for correction. First, through a skeletal key point detection algorithm (such as OpenPose), extract the data of the amplitude of body movements of the participating personnel in each interaction event cluster, including parameters such as the change in joint angles and the movement speed of the limbs. The preset action standard library stores the characteristic models of various normal and abnormal actions. For example, the range of joint angle changes for normal actions such as "waving" and "running", as well as the characteristic thresholds for abnormal actions such as "pushing and shoving" and "climbing". The system matches the detected body movement data with the models in the standard library and calculates the cosine similarity of the action feature vectors. If the similarity exceeds the preset matching threshold (such as 0.7), it is determined as an abnormal action pattern.

[0072] For the successfully matched abnormal action patterns, a weighted sum is generated according to their risk levels to obtain the action risk coefficient. The risk levels are divided into three levels: low risk (such as slight pushing and shoving, weight 0.3), medium risk (such as threatening with a weapon, weight 0.6), and high risk (such as weight 1.0). The calculation formula for the action risk coefficient is:

[0073] Among them, is the action risk coefficient, is the number of abnormal action patterns, is the th risk level weight of the abnormal action, is the th matching similarity of the abnormal action.

[0074] The action risk coefficient and the duration difference are linearly fused to obtain the corrected value of the interaction anomaly index, and the secondary abnormal area is determined based on the corrected value. The linear fusion formula is:

[0075] Among them, is the corrected interactive anomaly index, is the duration difference (unit: minute), is the fusion weight coefficient (value range 0 - 1, default value 0.5). Through this correction mechanism, the interactive anomaly index can more comprehensively reflect the risk of behavior. For example, if the duration difference of a certain interactive event cluster is 3 minutes and it contains 2 medium-risk abnormal actions (similarity degrees are 0.8 and 0.7 respectively), then the action risk coefficient , the corrected interactive anomaly index .

[0076] During the dynamic path adjustment process, the system real-time monitors the change rate of the direction deviation degree of high-risk behavior nodes. The change rate of the direction deviation degree is the ratio of the difference in the direction deviation degree within adjacent time intervals to the time interval, and the calculation formula is:

[0077] Among them, is the change rate of the direction deviation degree (unit: degree / second), is the direction deviation degree at the current moment, is the direction deviation degree at the previous moment, is the time interval (unit: second). The preset fluctuation threshold is an empirical parameter. For example, it is set to 5 degrees / second. When exceeds the preset fluctuation threshold, it indicates that the moving direction of the high-risk behavior node has changed violently, and there may be external interference or abnormal behavior escalation. At this time, the path backtracking mechanism is triggered.

[0078] After the path backtracking mechanism is started, the system extracts the scan records of group aggregation nodes and staying nodes within the previous time period (such as the past 10 minutes), including data such as the scan time, stay time, and aggregation risk coefficient of each node. Based on these historical data, the duration weights of the first scan sequence (group aggregation nodes) and the second scan sequence (staying nodes) are recalculated. The updated calculation method of the duration weight is: according to the abnormal event occurrence rate of each type of node in the historical scan, dynamically adjust its duration ratio. For example, if the proportion of abnormal events triggered by group aggregation nodes in the historical data is relatively high, then increase the duration weight of the first scan sequence to 70%, otherwise reduce it to 30%.

[0079] Interpolate and fuse the recalculated duration weights with the current duration weights of the first scan sequence and the second scan sequence to generate anti-interference new weights, and determine the scan order and scan duration with the anti-interference new weights. The interpolation fusion is implemented through a linear interpolation algorithm. For example, the recalculated duration weight is (the first scanning sequence) and (the second scanning sequence), and the duration weights of the current first scanning sequence and the second scanning sequence are respectively and respectively. Then the new weight is , , where is the interpolation coefficient (the value range is 0 to 1, and the default value is 0.5), and are the new weights. By fusing historical experience and current data, the scanning sequence has stronger anti-interference ability, avoiding misjudgment of the monitoring strategy caused by short-term data fluctuations.

[0080] At the system implementation level, the correction of the interaction anomaly index and the path backtracking mechanism are completed collaboratively by the real-time data processing module and the historical data query module. The skeleton key point detection and action matching algorithms run on the edge computing nodes to ensure real-time performance; the historical scanning records are stored in the time series database to support fast query and analysis. The monitoring personnel can view the correction process of the interaction anomaly index and the path backtracking records through the system log, which is convenient for post-event auditing and policy optimization.

[0081] It should be noted that in this article, relational terms such as first and second are only used 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 term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0082] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

[0083] Matters not described in the present invention apply to the prior art.

Claims

1. A video surveillance method for abnormal behaviors of community personnel based on human factors engineering, characterized in that, The method includes: Performing spatial partitioning on the community surveillance video data based on a preset activity area range to obtain the surveillance data of each sub-region, where the surveillance data of each sub-region includes personnel trajectory data and environmental interaction data; Determining the anomaly level of each sub-region according to the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data in each sub-region; Calculating the key surveillance nodes of each sub-region according to the anomaly levels, personnel trajectory data, and environmental interaction data of each sub-region, and determining the scanning order and scanning duration of the key surveillance nodes based on dynamic path adjustment; Generating observation anchor points for each sub-region at the boundary of each sub-region according to a preset observation angle, the key surveillance nodes of each sub-region, and spatial orientation information, and generating redundant anchor points for each sub-region at the boundary of each sub-region according to a preset redundant distance, the key surveillance nodes, and spatial orientation information; Associating and integrating the key surveillance nodes, observation anchor points, and redundant anchor points based on the scanning order and scanning duration of the key surveillance nodes to generate a video surveillance path for the community.

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

3. The method for video monitoring of abnormal behaviors of community personnel based on human factors engineering according to claim 1, characterized in that, Determining the anomaly level of each sub-region according to the movement frequency of the personnel trajectory data and the interaction intensity of the environmental interaction data in each sub-region, including: For the current sub-region, when the number of movements per unit time in the personnel trajectory data exceeds a preset frequency threshold, marking the current sub-region as a first-level anomaly region; When the number of movements per unit time does not exceed the preset frequency threshold, performing spatio-temporal clustering on the environmental interaction data based on a preset clustering radius to obtain each interaction event cluster; Recording the difference between the duration of each interaction event cluster and a preset reference duration as the duration difference, and calculating an interaction anomaly index according to the duration difference; If the interaction anomaly index exceeds a preset anomaly threshold, marking the current sub-region as a second-level anomaly region; If it does not exceed the preset anomaly threshold, marking the current sub-region as a regular surveillance region.

4. The method for video monitoring of abnormal behaviors of community personnel based on human factors engineering according to claim 3, characterized in that The surveillance data of each sub-region further includes wearable feature data and group aggregation data, and the key surveillance nodes include high-risk behavior nodes, stay nodes, and group aggregation nodes. Calculating the key surveillance nodes of each sub-region according to the anomaly levels, personnel trajectory data, and environmental interaction data of each sub-region, including: For the current sub-region, screening the group aggregation data based on a preset density threshold to obtain a high-density aggregation region; Calculating an aggregation risk coefficient according to the personnel distribution uniformity and movement direction consistency of the high-density aggregation region; 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; Extract the coordinate points in the personnel trajectory data where the stay time exceeds a preset duration, and mark them as the stay nodes in combination with the interaction types of the environmental interaction data.

5. The method for video monitoring of abnormal behaviors of community personnel based on human factors engineering according to claim 4, wherein, After calculating the key monitoring nodes, it 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 according to the spatio-temporal correlation of each candidate node in the candidate set of high-risk behavior nodes, and retaining the candidate nodes that meet the preset correlation strength; Performing vector matching on the spatial coordinates of the retained candidate nodes that meet the preset correlation strength and the moving direction of the personnel trajectory data, and screening out the candidate nodes with a direction deviation greater than a preset angle as the final high-risk behavior nodes.

6. The video surveillance method for abnormal behaviors of community personnel based on human factors engineering according to claim 5, wherein, The dynamic path adjustment includes: Sorting the group aggregation nodes in descending order of the aggregation risk coefficient to generate a first scan sequence; Sorting the stay nodes in descending order of the ratio of the stay time to the preset duration to generate a second scan sequence; Setting a total scan duration, the two scan sequences allocate the total scan duration according to the duration weights, dynamically adjusting the duration weights of the first scan sequence and the second scan sequence based on the average value of the difference between the direction deviation of all high-risk behavior nodes and the preset angle, updating the scan durations of the first scan sequence and the second scan sequence, and performing monitoring scans with the updated scan durations of the two scan sequences, and preferentially scanning the scan sequence with a larger duration weight to obtain the scan order and scan duration of the key monitoring nodes.

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

8. The method for video monitoring of abnormal behaviors of community personnel based on human factors engineering according to claim 3, characterized in that, The calculation of the interaction anomaly index further includes: Extracting the limb movement amplitude data of the participants in each interaction event cluster, and matching abnormal movement patterns based on a preset action standard library; Weighted summing the successfully matched abnormal movement patterns according to the danger level to generate an action risk coefficient; Linearly fusing the action risk coefficient with the duration difference to output a correction value of the interaction anomaly index.

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

10. A video monitoring method for abnormal behaviors of community personnel based on human factors engineering according to claim 1, characterized in that, The dynamic adjustment of the preset activity area range includes: Obtain the building layout change data in the community in real time, and generate a virtual space grid based on the changed topological structure; According to the difference value between the historical data and the real-time data of the personnel density distribution in the virtual space grid, dynamically expand or contract the boundary threshold of the activity area range; Superimpose and calculate the boundary threshold with the preset minimum safety distance to generate updated space division parameters.

Images