Abnormal behavior state recognition and early warning method based on video image processing

Abstract

The invention provides a behavior abnormal state recognition early warning method based on video image processing, and the method comprises the steps: collecting basic data, carrying out the scene adaptive feature extraction, and generating a scene adaptive behavior feature vector; performing multi-dimensional abnormal behavior energy evaluation and grading on the scene adaptive behavior feature vector to obtain an abnormal behavior energy grading matrix; performing dynamic resource optimization and multi-camera cooperative tracking based on the abnormal behavior energy grading matrix to obtain a cross-camera abnormal event association graph; and verifying the cross-camera abnormal event association diagram, generating graded early warning information, evaluating resource allocation efficiency, generating optimal configuration parameters, and feeding back the optimal configuration parameters to the resource optimization step. Through multi-dimensional abnormal energy evaluation, dynamic resource optimization and multi-camera cooperative tracking, the abnormal behavior recognition accuracy and the system resource utilization efficiency are improved.

Description

Technical Field

[0001] The invention relates to a behavior recognition method, in particular to a behavior abnormal state recognition and early warning method based on video image processing. Background Art

[0002] In places such as smart cities, important infrastructure, transportation hubs and large commercial complexes, real-time detection and warning of potential abnormal behavior can effectively prevent security incidents and minimize casualties and property losses.

[0003] Existing abnormal behavior recognition technologies are mainly divided into two categories: rule-based methods and statistical learning-based methods. Rule-based methods identify anomalies by pre-defining a series of behavior patterns and judgment criteria, such as motion speed detection with fixed thresholds, trajectory deviation judgment, and regional intrusion detection. This type of method is simple to implement and has high computational efficiency, but lacks adaptability to complex scenarios and the rule setting is too rigid. Statistical learning-based methods identify behaviors that deviate from normal patterns as abnormalities by building statistical models of normal behaviors, such as Gaussian mixture models, autoencoders, and single-class support vector machines. This type of method performs well in simple scenarios, but lacks robustness for changing environments. In addition, most existing systems use fixed sampling rates and uniform resource allocation strategies, lacking differentiated processing of the importance of different scenarios and events. Traditional multi-camera systems usually use a simple neighborhood information sharing mechanism, with each camera node working independently, lacking effective coordination mechanisms and resource optimization strategies.

[0004] The core problem facing current abnormal behavior recognition technology is the singleness of the abnormal assessment model and the static nature of resource allocation. In terms of abnormal assessment, traditional methods often rely solely on the spatial distance measurement of video images (such as Mahalanobis distance or Euclidean distance) to assess the degree of abnormal behavior, ignoring the changing characteristics of the time dimension and the behavioral context information, making it difficult for the system to distinguish between temporary deviations and persistent anomalies, and unable to capture the local aggregation effect and evolution trend of abnormal behavior. Other information, such as activity information in the form of text and voice, is also not integrated.

[0005] This one-dimensional evaluation leads to a high false alarm rate, especially in crowded or complex environments. In terms of resource allocation, most existing systems use preset, static resource allocation strategies, ignoring the dynamics of abnormal events and the differentiated processing requirements. Relying solely on video image processing, the resource occupancy rate is high. When the system faces multiple abnormal events occurring simultaneously, key abnormal events may be delayed due to insufficient resources, resulting in inefficient use and waste of computing resources. Summary of the invention

[0006] The purpose of the invention is to provide a method for identifying and warning of abnormal behavior states based on video image processing, which improves the video classification and processing speed by combining multimodal information such as activity information, user setting data, activity information, etc., and then quickly screens out abnormal areas, and then allocates more resources to abnormal areas, and then quickly identifies abnormal states, in order to solve at least one technical problem existing in the prior art.

[0007] The technical solution is a method for identifying and warning abnormal behavior based on video image processing, comprising the following steps:

[0008] Collect basic data and perform scene-adaptive feature extraction based on it to generate scene-adaptive behavior feature vectors; basic data includes original video stream data, user setting data and activity information data;

[0009] Perform multi-dimensional abnormal behavior energy evaluation and classification on the scene-adaptive behavior feature vector to obtain an abnormal behavior energy classification matrix;

[0010] Based on the abnormal behavior energy classification matrix, dynamic resource optimization and multi-camera collaborative tracking are performed to obtain a cross-camera abnormal event correlation graph;

[0011] Verify the cross-camera abnormal event correlation graph and generate graded warning information.

[0012] In another embodiment of the present application, it also includes evaluating resource allocation efficiency, generating optimized configuration parameters and feeding back to the resource optimization step.

[0013] According to one aspect of the present application, the steps of performing multi-dimensional abnormal behavior energy evaluation and classification to obtain an abnormal behavior energy classification matrix include:

[0014] Based on the historical data of scene-adaptive behavior feature vectors, a time-sensitive background model with memory decay characteristics is constructed, normal behavior boundaries are established, and a typical abnormal pattern library is extracted;

[0015] Based on the typical abnormal pattern library, an abnormal energy function is constructed; for the adaptive behavior feature vector of the current scenario, the abnormal energy function is applied to calculate the abnormal energy value including spatial distance, temporal distance, complexity coefficient and business impact factor, and generate abnormal behavior indicators;

[0016] Establish a regional value map, conduct a multi-dimensional comprehensive assessment based on abnormal behavior indicators, classify abnormal behaviors, and generate an abnormal behavior energy classification matrix.

[0017] According to one aspect of the present application, the step of applying an abnormal energy function to calculate an abnormal energy value including a spatial distance, a temporal distance, a complexity coefficient, and a business impact factor to generate an abnormal behavior indicator includes:

[0018] Construct the abnormal energy function E(b)=α·Ds(b,M)+β·Dt(b,H)+γ·C(b)+δ·I(b);

[0019] Where Ds(b, M) represents the spatial distance between behavior b and the normal behavior boundary M, Dt(b, H) represents the temporal distance between behavior b and the historical behavior sequence H, C(b) represents the complexity coefficient of behavior b, I(b) represents the business impact factor of behavior b, and α, β, γ, and δ are adjustable weight parameters;

[0020] Applying the abnormal energy function to the current scene adaptive behavior feature vector to calculate the abnormal energy value;

[0021] Based on the abnormal energy value and the preset judgment criteria, the degree of abnormal behavior is determined and an abnormal behavior indicator is generated.

[0022] According to one aspect of the present application, the steps of establishing a regional value map, combining abnormal behavior indicators to conduct a multi-dimensional comprehensive assessment, grading abnormal behaviors, and generating an abnormal behavior energy grading matrix include:

[0023] Based on the business importance of the preset monitoring areas, a regional value map is constructed to show the relative value of each area;

[0024] The abnormal behavior index and the regional value map are weighted and combined to construct a weighted abnormal importance index that reflects the importance of abnormal behavior;

[0025] Applying the hierarchical analysis algorithm, a multi-dimensional comprehensive evaluation of the weighted abnormal importance index is performed according to the pre-stored abnormal behavior type, occurrence location and impact range to obtain the evaluation result;

[0026] Based on the evaluation results, abnormal behaviors are divided into five levels: critical, high-risk, medium-risk, low-risk, and prompt, and an abnormal behavior energy classification matrix is ​​generated.

[0027] According to one aspect of the present application, the weight parameters α, β, γ, δ in the abnormal energy function are dynamically adjusted by the following steps:

[0028] Collect historical data containing confirmed abnormal behaviors and extract historical anomaly detection results;

[0029] Based on the historical anomaly detection results, the particle swarm optimization algorithm is applied to calculate the weight parameter combination that maximizes the detection accuracy;

[0030] Update the weight parameters α, β, γ, δ in the anomaly energy function, maintaining the constraint condition of α+β+γ+δ=1;

[0031] The weight parameters are adjusted periodically according to the current scene changes to make the abnormal energy function adapt to the characteristics of different scenes.

[0032] According to one aspect of the present application, after calculating the abnormal energy value, the step of calculating the local energy density is also included:

[0033] Construct a local energy density function ρ(b, r) to represent the abnormal energy density within a radius r around behavior b; ρ(b, r) = ∑E(bi) w(d(b, bi)), where bi is other behaviors within the radius r, E(bi) is its abnormal energy value, d(b, bi) is the distance between behaviors b and bi, and w is a weight function based on distance;

[0034] Gaussian kernel function is used as the weight function w(d)=exp(-d 2 / 2σ 2 ), where σ is an adjustable influence range parameter and d is the distance;

[0035] The local energy density values ​​of behavior b are calculated in aggregate to capture the local aggregation effect of abnormal behaviors.

[0036] According to one aspect of the present application, after calculating the abnormal energy value, the step of constructing an energy fluctuation function to analyze the time series energy feature vector is also included:

[0037] Construct the energy fluctuation function Φ(b, t)=[Δe1, Δe2, ..., Δe n , σ, τ], which is used to analyze the energy variation pattern of behavior b within the time window t;

[0038] Calculate the rate of energy change Δe at consecutive time points i =(E(b, t i )-E(b,t i₋1 )) / E(b,t i₋1 ), generating energy change sequence;

[0039] Calculate the fluctuation characteristics of the energy change sequence, including the standard deviation σ and the autocorrelation coefficient τ, to distinguish stable, fluctuating and mutation anomalies;

[0040] Combining the energy change sequence and fluctuation characteristics, a time series energy feature vector is generated to characterize the time series development characteristics of abnormal behavior.

[0041] According to one aspect of the present application, the step of constructing a fusion enhanced energy function is further included:

[0042] Construct the fusion enhanced energy function E'(b)=E(b)·[1+λ1·f(ρ(b, r))+λ2·g(Φ(b, t))], where E(b) is the original abnormal energy value, ρ(b, r) is the local energy density value, Φ(b, t) is the time series energy feature vector, and λ1 and λ2 are adaptive weight coefficients;

[0043] Define nonlinear mapping functions f(x)=tanh(k1·x) and g(y)=max(0,1-exp(-k2·||y||)), where k1 and k2 are adjustment coefficients used to convert density values ​​and time series features; x and y represent different variables;

[0044] The fusion enhanced energy function is applied to calculate the enhanced anomaly energy index to achieve differentiated evaluation of isolated anomalies, clustered anomalies and evolving anomalies.

[0045] According to one aspect of the present application, the steps of performing dynamic resource optimization and multi-camera collaborative tracking to obtain a cross-camera abnormal event association graph include:

[0046] Based on the abnormal behavior energy classification matrix, the pre-configured resource demand prediction model is applied to calculate the predicted resource demand value of each camera node and generate a global resource demand matrix;

[0047] Apply the pre-configured resource allocation optimization model to solve the global resource demand matrix, consider the detection priority and system load balancing, and generate a camera resource allocation plan;

[0048] Based on the abnormal behavior energy classification matrix and camera resource allocation scheme, the target of attention is determined, a multi-camera collaborative tracking protocol is constructed and combined with the pre-stored camera topology map for collaborative tracking, and a cross-camera abnormal event correlation map is generated.

[0049] According to one aspect of the present application, the steps of verifying the cross-camera abnormal event association graph and generating graded warning information, evaluating resource allocation efficiency, generating optimized configuration parameters and feeding back to the resource optimization step include:

[0050] Apply spatiotemporal consistency verification to the cross-camera abnormal event association graph to generate verified abnormal events, and generate abnormal event impact range graphs and graded warning information based on their energy levels and spatial distribution;

[0051] Collect system operation data to form a system performance indicator set, apply the pre-configured resource benefit evaluation model to calculate the resource benefit indicator, optimize the resource allocation weight parameters, and generate the optimized configuration parameters;

[0052] The optimized configuration parameters are fed back to the resource demand prediction model and the resource allocation optimization model, and the pre-configured typical abnormal pattern library and feature extraction configuration parameters are updated to perform closed-loop optimization.

[0053] Beneficial effect: The present invention overcomes the problems of lack of scene adaptability, single evaluation and static resource allocation of traditional video image processing systems by integrating multi-source data, including video image processing of activity information data, and achieves improved accuracy of abnormal behavior detection and optimization of system resource utilization efficiency, improves abnormal behavior detection rate while reducing false alarm rate, and improves the practicality and reliability of video surveillance systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flowchart of the steps of a method for identifying and warning of abnormal behavior states based on video image processing provided in an embodiment of the present application.

[0055] Figure 2 A flowchart of the steps for obtaining an abnormal behavior energy grading matrix provided in an embodiment of the present application.

[0056] Figure 3 A flowchart of the steps for generating an abnormal behavior energy grading matrix provided in an embodiment of the present application.

[0057] Figure 4 A flowchart of the steps for obtaining a cross-camera abnormal event correlation diagram provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. 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 creative work should fall within the scope of protection of the present invention.

[0059] It should be noted that in order to clearly show the steps of this application, serial numbers are marked for each step in the specification. These serial numbers are only used for the convenience of explanation and do not limit the order of execution of the steps. In actual operation, according to the technical requirements of the specific implementation scenario, the steps can be executed in a different order than that shown in the specification, and in some cases, parallel processing between steps can be achieved.

[0060] The present invention is described below in conjunction with preferred implementation steps. Figure 1 As shown, a method for identifying and warning abnormal behavior states based on video image processing includes the following steps:

[0061] S1. Collect basic data and perform scene-adaptive feature extraction based on it to generate scene-adaptive behavior feature vectors; the basic data includes original video stream data, user setting data and activity information data collected by the distributed monitoring network.

[0062] Specifically, the raw video stream data can be activities in crowded areas, key facilities, safe passages, etc., as well as video frames, timestamp information, and related metadata. User-defined data can be the criteria for determining abnormal behavior (such as thresholds for certain behaviors), the priority of specific monitoring areas, and the camera field of view and resolution requirements. Activity information data can be specific activities or event information detected in the target area, such as: the trajectory, speed, and behavior pattern of personnel activities; situations where groups gather or suddenly disperse; abnormal activities, such as outliers, intrusions, etc. These data are pre-processed by noise reduction, resolution unification, etc. to generate standardized video frames, and then scene-adaptive behavior feature vectors are adaptively extracted according to the scene type. The scene-adaptive behavior feature vector is a set of data describing behavior characteristics extracted according to a specific scene, such as the movements of personnel, activity distribution, etc.

[0063] S2. Perform multi-dimensional abnormal behavior energy evaluation and classification on the scene-adaptive behavior feature vector to obtain an abnormal behavior energy classification matrix.

[0064] Specifically, the abnormal behavior energy index is calculated based on the scenario-adaptive behavior feature vector, and then multi-dimensional classification is performed according to the business importance and abnormality degree, and the abnormal behavior energy classification matrix is ​​output. In other words, by analyzing the scenario-adaptive behavior feature vector, the behavior is evaluated from multiple dimensions to determine whether it is abnormal, which may include time, space, or behavior intensity. The evaluation results are classified or layered, and they are divided into different levels according to the severity or characteristics of the abnormal behavior.

[0065] S3, based on the abnormal behavior energy classification matrix, dynamic resource optimization and multi-camera collaborative tracking are performed to obtain a cross-camera abnormal event correlation graph;

[0066] Specifically, the computing resource allocation is estimated and optimized according to the abnormal behavior energy classification matrix, the camera resource allocation plan is generated and multi-camera collaborative tracking is implemented, and the cross-camera abnormal event association graph is constructed and output. In other words, the available resources are intelligently adjusted and allocated according to the classification information in the matrix. For example, more computing power or higher monitoring accuracy can be allocated to high-priority anomalies. Different cameras work together to track one or more targets that may have abnormal behavior. For example, when a target leaves the field of view of one camera, another camera will take over the tracking. The tracking information from multiple cameras is combined to form a graphical result. This kind of chart shows the relationship between different abnormal events and the dynamic connection across cameras.

[0067] S4. Verify the cross-camera abnormal event correlation diagram and generate graded warning information. At the same time, evaluate the resource allocation efficiency and abnormality detection rate, generate optimized configuration parameters and feed them back to the resource optimization step.

[0068] Specifically, check whether the cross-camera abnormal event correlation graph is accurate to ensure the reliability of abnormal events and their relationships. Based on the verification results, classify the abnormal events by severity and generate corresponding warning information. For example, severe abnormalities may trigger high-level alarms, while minor abnormalities may only be recorded or prompted. Check whether the current use of resources (such as computing power and monitoring equipment) is reasonable and evaluate the optimization effect. Measure the system's ability to identify abnormalities and ensure that the underreporting and false alarm rates are within an acceptable range. Based on the above evaluation results, generate new system configuration parameters for further optimizing resource allocation and anomaly detection processes. Return the new configuration parameters to the resource optimization link and adjust the system to continuously improve monitoring efficiency and abnormality handling capabilities.

[0069] This embodiment achieves the improvement of abnormal behavior detection accuracy and optimization of system resource utilization efficiency by integrating a complete closed-loop solution of multi-source video acquisition, feature extraction, abnormal behavior energy assessment, resource optimization and multi-camera collaborative tracking, as well as warning generation and strategy optimization. The organic combination of scene adaptation, multi-dimensional evaluation and dynamic resource scheduling overcomes the problems of traditional systems' lack of scene adaptability, single evaluation and static resource allocation. It can accurately identify various abnormal behaviors and issue graded warnings in a complex and changeable monitoring environment. Especially in large places such as airports, stations and other crowded areas, this embodiment can increase the abnormal behavior detection rate from the traditional 75% to more than 92%, while reducing the false alarm rate from 25% to less than 8%, improving the practicality and reliability of the video surveillance system.

[0070] In order to achieve more accurate abnormal state recognition, the venue activity information can be integrated as context data while collecting raw video stream data from the distributed monitoring network. Specifically:

[0071] Introduce event information data source: Establish an event information database, including the time, location, type, expected flow of people and other information of planned events such as mall promotions, performances, maintenance work, etc. This data can be obtained through the management system API interface, or entered by operators through a dedicated interface. The details are as follows:

[0072] Fusion of activity information and video data: In the scene adaptive feature extraction stage, the activity information is used as context features and fused with the video features. For example, associate a corresponding activity label A = {a1, a2, ..., am} with each monitoring area, where ai represents the activity type of the area in a specific time period; define the expected behavior pattern template T = {t1, t2, ..., tn} for each activity type, such as crowd density, mobility and other features corresponding to "promotion activities";

[0073] Generate context-enhanced feature vector: expand the dimension of the scene-adaptive behavior feature vector and add activity-related features: B' = [B, A_feat, C_score]; where B is the original behavior feature vector; A_feat is the current area activity feature; C_score is the consistency score between the behavior and the expected pattern of the current activity.

[0074] Construct activity-aware anomaly energy function: modify the anomaly energy function and add the activity context adjustment factor: E(B') = E(B) × ω(A, B); where ω(A, B) is the activity context adjustment factor: when behavior B is highly consistent with the current activity A, the ω value is small (such as 0.3), reducing the anomaly energy; when there is no related activity or the behavior does not meet the activity expectations, the ω value is close to 1, maintaining the original anomaly energy; E(B) is the original anomaly energy function.

[0075] Perform dynamic threshold activity adaptation: Dynamically adjust the abnormal threshold according to the activity type and scale: θi(t, A) = θi(t) × φ(A); where φ(A) is the activity type adjustment coefficient, for example: large-scale promotional activities: φ=1.5, increase the abnormal judgment threshold, and tolerate higher crowd density; routine operations: φ=1.0, maintain the standard threshold; store closure for maintenance: φ=0.7, lower the abnormal judgment threshold, and be more sensitive to any personnel activities.

[0076] That is to say, according to one aspect of the present application, step S1 may also be: collecting original video stream data and venue activity information from a distributed monitoring network, performing data fusion and scene adaptive feature extraction, and generating a context-enhanced scene adaptive behavior feature vector.

[0077] Specifically, the original video stream data is collected and processed for timestamp synchronization, noise reduction and resolution unification to generate a standardized video frame sequence; at the same time, the activity data of the current monitoring area is obtained from the activity information database, including activity type, time range and expected crowd flow parameters, to generate spatiotemporal-correlated activity context information. A comprehensive scene analysis is performed on the standardized video frame sequence combined with the activity context information, and a context-aware feature extraction strategy selector is constructed to generate scene-activity composite type labels and feature extraction configuration parameters. According to the scene-activity composite type labels and feature extraction configuration parameters, the corresponding feature extraction algorithm is applied to the standardized video frame sequence to extract the multimodal original feature set. Through dimensionality reduction and fusion, combined with the temporal change pattern analysis and the activity expectation pattern matching degree calculation, a context-enhanced scene-adaptive behavior feature vector is generated.

[0078] In this way, the system can distinguish between crowds caused by normal shopping mall activities and truly abnormal crowds. For example, a high-density crowd in a promotion area will not be mistakenly judged as abnormal, while a crowd of the same density appearing in an emergency passage will be correctly identified as abnormal. This improvement will reduce the system's false alarm rate, especially in places such as shopping malls and exhibition halls with frequent activities, while maintaining a high sensitivity to real abnormal situations. In addition, activity information can also optimize resource allocation strategies. The system can allocate more resources to planned activity areas in advance to ensure that efficient anomaly detection capabilities are maintained when the flow of people increases. This active resource scheduling based on prior knowledge will further improve the overall effectiveness of the system.

[0079] According to another aspect of the present application, the step of performing scene adaptive feature extraction and generating a scene adaptive behavior feature vector includes:

[0080] S11, collect and perform timestamp synchronization, noise reduction and resolution unification processing on basic data to generate a standardized video frame sequence;

[0081] S12, performing scene analysis on the standardized video frame sequence, constructing a feature extraction strategy selector, and generating scene type labels and feature extraction configuration parameters;

[0082] S13. According to the scene type label and feature extraction configuration parameters, the corresponding feature extraction algorithm is applied to the standardized video frame sequence to extract the multimodal original feature set, and the scene-adaptive behavior feature vector is generated through dimensionality reduction and fusion combined with temporal change pattern analysis.

[0083] Specifically, basic data collection and preprocessing are performed. Raw video stream data, user setting data, and activity information data are collected from multiple camera nodes in the distributed monitoring network; timestamp synchronization is performed to eliminate the time deviation between different cameras to obtain a time-synchronized video stream; adaptive noise reduction is performed on the time-synchronized video stream, and the optimal filter is selected according to the noise distribution characteristics to generate a noise-reduced video stream; the resolution of the noise-reduced video stream is unified and the frame rate is standardized to generate a standardized video frame sequence.

[0084] Generate scene self-learning classification and feature extraction strategies. Perform scene analysis on standardized video frame sequences to extract scene feature descriptors; based on scene feature descriptors, use clustering algorithms (such as density-weighted K-means) to automatically classify scenes and obtain scene type labels; build a decision tree-based feature extraction strategy selector to match the optimal feature extraction method combination for each scene type label; generate feature extraction configuration parameters, including the optimal feature extraction algorithm parameter settings for various scenes.

[0085] Perform adaptive spatiotemporal feature extraction and behavior characterization. According to the scene type label and feature extraction configuration parameters, apply the corresponding feature extraction algorithm to the standardized video frame sequence; extract the multimodal original feature set including optical flow features, trajectory features, posture features, etc.; apply the sparse coding algorithm to reduce the dimension and fuse the multimodal original feature set to obtain a compressed feature vector; use the dynamic time window mechanism to analyze the temporal change pattern of the compressed feature vector and construct a scene-adaptive behavior feature vector.

[0086] This embodiment solves the problem that traditional fixed feature extraction methods are difficult to adapt to diversified monitoring scenarios. It realizes automatic recognition of different scene types and dynamic selection of the optimal feature extraction algorithm, so that the system can adaptively extract the most discernible features for different scenes such as shopping malls, streets, and campuses. For example, in a crowded shopping mall scene, the system will give priority to extracting group movement patterns and individual trajectory deviation features; while in a campus environment, it focuses on extracting personnel retention and abnormal interaction features. After testing, compared with the fixed feature extraction method, this embodiment has improved the feature expression capability in cross-scene applications by 37%, laying a solid foundation for subsequent abnormal behavior identification.

[0087] like Figure 2 As shown, according to one aspect of the present application, the steps of performing multi-dimensional abnormal behavior energy evaluation and classification and obtaining an abnormal behavior energy classification matrix include:

[0088] S21. Based on the historical data of the scene-adaptive behavior feature vector, a time-sensitive background model with memory decay characteristics is constructed, normal behavior boundaries are established, and a typical abnormal pattern library is extracted;

[0089] S22. Based on the typical abnormal pattern library, an abnormal energy function is constructed; for the current scenario adaptive behavior feature vector, the abnormal energy function is applied to calculate the abnormal energy value including spatial distance, temporal distance, complexity coefficient and business impact factor, and generate an abnormal behavior indicator;

[0090] S23. Establish a regional value map, conduct a multi-dimensional comprehensive assessment based on abnormal behavior indicators, classify abnormal behaviors, and generate an abnormal behavior energy classification matrix.

[0091] Specifically, dynamic background and normal behavior patterns are constructed. Based on the historical data of scene-adaptive behavior feature vectors, a Gaussian mixture model is used to construct a dynamic background model, and a time-weighted coefficient is applied to the dynamic background model to form a time-sensitive background model with memory decay characteristics; a single-class support vector machine is used to establish normal behavior boundaries and define regular behavior patterns in the scene; and a typical abnormal pattern library is extracted and stored from historical data through an outlier analysis algorithm.

[0092] This embodiment effectively solves the problem that traditional anomaly detection systems are insensitive to time changes and lack business relevance. The system can simultaneously consider the spatial abnormality, time evolution characteristics and business importance of the behavior, and classify abnormal behaviors into five levels of fine classification. In practical applications, this embodiment can accurately distinguish between temporary deviations and persistent anomalies, and effectively reduce the false alarm rate caused by normal behaviors such as short stops or changes in direction. Tests show that the false alarm rate is reduced by 67% in complex environments. At the same time, by combining regional value maps, the system can give higher priority to abnormal behaviors in high-value areas (such as safety exits, equipment rooms, etc.), thereby improving the response speed to potential security threats.

[0093] According to one aspect of the present application, the step of generating an abnormal behavior indicator includes:

[0094] Construct the abnormal energy function E(b)=α·Ds(b,M)+β·Dt(b,H)+γ·C(b)+δ·I(b);

[0095] Where Ds(b, M) represents the spatial distance between behavior b and the normal behavior boundary M, Dt(b, H) represents the temporal distance between behavior b and the historical behavior sequence H, C(b) represents the complexity coefficient of behavior b, I(b) represents the business impact factor of behavior b, α, β, γ, δ are adjustable weight parameters; E(b) is the abnormal energy value of behavior b;

[0096] Applying the abnormal energy function to the current scene adaptive behavior feature vector to calculate the abnormal energy value;

[0097] Based on the abnormal energy value and the preset judgment criteria, the degree of abnormal behavior is determined and an abnormal behavior indicator is generated.

[0098] This embodiment solves the problem of single dimension of traditional anomaly assessment methods. The anomaly energy function integrates spatial anomaly (such as trajectory deviation), temporal anomaly (such as behavior duration), behavior complexity (such as the frequency of changes in movement direction) and business impact (such as proximity to important facilities) into a unified anomaly energy value, achieving a comprehensive evaluation of abnormal behavior. Experiments have shown that compared with traditional methods based only on spatial distance, this embodiment has improved the recognition accuracy by 26% in complex scenarios, especially for complex abnormal behaviors such as "slowly approaching sensitive areas" and "repeatedly wandering". In actual deployment at transportation hubs, this embodiment successfully identified 93% of potential security threats that could not be detected by traditional methods, effectively improving the practicality of the system.

[0099] like Figure 3 As shown, according to one aspect of the present application, the step of generating an abnormal behavior energy grading matrix includes:

[0100] S231, constructing a regional value map indicating the relative value of each region based on the business importance of the preset monitoring region;

[0101] S232, performing weighted combination of the abnormal behavior index and the regional value map to construct a weighted abnormal importance index reflecting the importance of the abnormal behavior;

[0102] S233, applying a hierarchical analysis algorithm, performing a multi-dimensional comprehensive evaluation on the weighted abnormal importance index according to the pre-stored abnormal behavior type, occurrence location, and impact range, to obtain an evaluation result;

[0103] S234. Based on the evaluation results, the abnormal behaviors are divided into five levels: critical, high risk, medium risk, low risk, and warning, and an abnormal behavior energy classification matrix is ​​generated.

[0104] This embodiment solves the problem of traditional abnormal behavior assessment ignoring business value by establishing a regional value map, constructing a weighted abnormal importance index, and applying a hierarchical analysis algorithm for comprehensive evaluation. The system can combine the business importance of the monitored area with the abnormal behavior index and conduct differentiated evaluations on abnormal behaviors in different areas. For example, when deployed in a financial institution, the system will assign different weights to abnormal behaviors in the ATM area, safe area, and ordinary business area to ensure that the security of high-value areas is prioritized. Test data shows that this embodiment can increase the detection priority of abnormal behaviors in important areas by 2.5 times, allowing security personnel to focus on handling the most threatening abnormal events. At the same time, the five-level classification mechanism (critical, high-risk, medium-risk, low-risk, and prompt) makes subsequent warning information more targeted, and improves the security response efficiency by 36% compared to binary classification (normal / abnormal).

[0105] According to one aspect of the present application, the weight parameters α, β, γ, δ in the abnormal energy function are dynamically adjusted by the following steps:

[0106] Collect historical data containing confirmed abnormal behaviors and extract historical anomaly detection results;

[0107] Based on the historical anomaly detection results, the particle swarm optimization algorithm is applied to calculate the weight parameter combination that maximizes the detection accuracy;

[0108] Update the weight parameters α, β, γ, δ in the anomaly energy function, maintaining the constraint condition of α+β+γ+δ=1;

[0109] The weight parameters are adjusted periodically according to the current scene changes to make the abnormal energy function adapt to the characteristics of different scenes.

[0110] This embodiment solves the problem that traditional fixed weights cannot adapt to different monitoring scenarios by dynamically adjusting the weight parameters in the abnormal energy function based on historical data. The particle swarm optimization algorithm is used to automatically find the weight parameter combination that maximizes the detection accuracy by analyzing the historical data of confirmed abnormal behaviors, while maintaining the constraint condition of α+β+γ+δ=1. In multi-scenario testing, this embodiment improves the adaptability of the system in different environments - in crowded areas such as stations, the system automatically increases the spatial distance weight α and the time distance weight β to pay attention to abnormal movement and retention of personnel; in relatively closed environments such as schools, the system increases the complexity coefficient weight γ and the business impact factor weight δ to pay attention to behavioral complexity and sensitive area contact. Actual measurements show that compared with fixed weights, this embodiment improves the average detection accuracy of the system in different scenarios by 23%, and can adapt to scene changes without manual intervention, reducing system maintenance and tuning costs.

[0111] According to one aspect of the present application, after calculating the abnormal energy value, the step of calculating the local energy density is also included:

[0112] Construct a local energy density function ρ(b, r) to represent the abnormal energy density within a radius r around behavior b; ρ(b, r) = ∑E(bi)·w(d(b, bi)), j∈{j | d(B, B j ) ≤ r}; where bi is other behaviors within radius r, E(bi) is its abnormal energy value, d(b,bi) is the distance between behaviors b and bi, and w is a weight function based on distance;

[0113] Gaussian kernel function is used as the weight function w(d)=exp(-d 2 / 2σ 2 ), where σ = r / 3 is an adjustable influence range parameter, r = 0.2 is the search radius in the standardized feature space, and d is the Euclidean distance in the feature space;

[0114] The local energy density values ​​of behavior b are calculated in aggregate to capture the local aggregation effect of abnormal behaviors.

[0115] This embodiment solves the problem that traditional anomaly detection methods cannot capture the spatial aggregation effect of abnormal behaviors by designing a local energy density function and using a Gaussian kernel function as a weight function. The system is able to evaluate the abnormal energy distribution within a radius r around the behavior, thereby identifying the spatial correlation pattern of group anomalies and abnormal behaviors. In practical applications, this embodiment can effectively detect group abnormal behaviors such as multiple people watching, gathering and wandering, and "chain reactions" caused by abnormal behaviors of a single person. Compared with the traditional method of independently evaluating each behavior, the accuracy rate in group anomaly identification is improved by 46%, especially in crowded places such as station squares, concert venues and other environments. This embodiment can detect potential safety risks caused by crowd gathering early, and the response time is advanced by an average of 3.7 seconds, which reserves more intervention time for security personnel and effectively prevents the occurrence of group safety incidents such as trampling.

[0116] According to one aspect of the present application, after calculating the abnormal energy value, the step of constructing an energy fluctuation function to analyze the time series energy feature vector is also included:

[0117] Construct the energy fluctuation function Φ(b, t)=[Δe1, Δe2, ..., Δe n , σ, τ], which is used to analyze the energy variation pattern of behavior b within the time window t;

[0118] Calculate the rate of energy change Δe at consecutive time points i =(E(b, t i )-E(b,t i₋1 )) / E(b,t i₋1 ), generating energy change sequence;

[0119] Calculate the fluctuation characteristics of the energy change sequence, including the standard deviation σ and the autocorrelation coefficient τ (lag=2), to distinguish stable, fluctuating and sudden abnormalities;

[0120] Combining the energy change sequence and fluctuation characteristics, a time series energy feature vector is generated to characterize the time series development characteristics of abnormal behavior, including the development trend information of the anomaly.

[0121] Specifically, according to the values ​​of the standard deviation σ and the autocorrelation coefficient τ, the anomalies are classified as follows: σ < 0.1 and |τ| < 0.2: stable anomalies; σ > 0.3 and |τ| > 0.5: fluctuating anomalies; σ > 0.5 and |τ| < 0.2: mutation anomalies.

[0122] This embodiment solves the problem that traditional anomaly detection methods cannot capture the time evolution characteristics of behavior by designing an energy fluctuation function to analyze the time series energy changes of behavior. By calculating the energy change rate, change sequence standard deviation and autocorrelation coefficient of continuous time points, a feature vector characterizing the time series development pattern of abnormal behavior is constructed. This enables the system to distinguish different types of stable anomalies (such as continuous retention), fluctuating anomalies (such as repeated attempts) and mutation anomalies (such as sudden running), and achieves a fine characterization of the time dimension of abnormal behavior. In the test, the recognition rate of progressive abnormal behavior (such as slow approach to restricted areas, gradual acceleration, etc.) in this embodiment was increased from 51% of the traditional method to 87%, and an early warning was issued 2.8 seconds in advance on average. In particular, for places with high security levels such as airports, nuclear power plants, etc., abnormal behavior can be identified as soon as it begins to appear, which improves the effectiveness of preventive safety measures.

[0123] According to one aspect of the present application, the step of constructing a fusion-enhanced energy function is also included:

[0124] Construct a fusion enhanced energy function E'(b)=E(b)·[1+λ1·f(ρ(b, r))+λ2·g(Φ(b, t))], where E(b) is the original abnormal energy value, ρ(b, r) is the local energy density value, Φ(b, t) is the time series energy feature vector, λ1 and λ2 are adaptive weight coefficients, λ1=0.3, λ2=0.4;

[0125] Define nonlinear mapping functions f(x)=tanh(k1·x) and g(y)=max(0,1-exp(-k2·||y||)), where k1 and k2 are adjustment coefficients, k1=2.0, k2=1.5, used to convert density values ​​and time series features; x and y represent different variables, |||| is the L2 norm;

[0126] By applying the fusion enhanced energy function, the enhanced anomaly energy index is calculated to achieve differentiated evaluation of isolated anomalies, aggregated anomalies and evolving anomalies. By fusion enhanced energy function, the original anomaly energy will be enhanced or suppressed according to the local density and timing characteristics, fusion enhanced energy function.

[0127] This embodiment solves the problem of incoordination of multiple anomaly assessment mechanisms operating independently by designing a fusion-enhanced energy function to organically integrate the original anomaly energy, local density and timing characteristics. Through the nonlinear mapping function, the assessment results of different dimensions are effectively converted and fused. The fusion-enhanced energy function enables the system to differentially evaluate isolated anomalies, clustered anomalies and evolving anomalies, and provide the most suitable energy enhancement or suppression for different types of anomalies. In comprehensive tests, compared with a single evaluation method, the fusion mechanism increased the accuracy of anomaly detection by 31%, especially reducing the false alarm rate by 43% in complex environments. In the actual application of financial institution security systems, this embodiment successfully identified slow-approaching anomalies and collaborative anomalies that traditional systems could not detect, effectively preventing multiple potential security incidents, and the value of the system was highly recognized by security personnel.

[0128] According to one aspect of the present application, the step of constructing a dynamic threshold function is also included:

[0129] Construct a dynamic threshold function θ for different levels of abnormalities i (t) = θ i，base ·[1+μ·V(t)-ν·S(t)], where θ i，base is the basic threshold of level i anomaly, set to [0.8, 0.6, 0.4, 0.2, 0.1]; V(t) is the scene activity, S(t) is the system stability index, μ and ν are adjustment parameters, μ=0.2, ν=0.3;

[0130] The scene activity V(t) is calculated based on the number and speed of moving targets in the monitoring area, and the system stability index S(t) is calculated based on the system resource occupancy rate and processing delay;

[0131] Apply the dynamic threshold function to generate adaptive grading thresholds for the five levels of critical, high risk, medium risk, low risk, and warning; and realize adaptive grading of abnormal behaviors under different scene states.

[0132] The enhanced abnormal energy index is compared with the dynamic classification threshold to generate a final abnormal behavior energy classification matrix.

[0133] Specifically, the scene activity V(t) = min(1.0, (N_obj / N_base) × (v_avg / v_base)); where N_base = 20, v_base = 1.0m / s. System stability index S(t) = min(1.0, (CPU_usage / 90%) ×(delay / delay_max)). According to the comparison between the enhanced abnormal energy value E'(b) and the dynamic threshold θᵢ(t), the abnormal behavior is divided into: E'(b) ≥ θ1(t): critical level (red); θ2(t) ≤ E'(b) < θ1(t): high risk level (orange); θ3(t) ≤ E'(b) < θ2(t): medium risk level (yellow); θ4(t) ≤ E'(b) < θ3(t): low risk level (blue); θ5(t) ≤ E'(b) < θ4(t): warning level (green). Finally, the abnormal behavior energy classification matrix EM is generated, which contains the abnormal target ID, location, abnormal energy value and classification information detected in each camera area.

[0134] This embodiment solves the problem that traditional fixed thresholds are difficult to adapt to scene changes and system load fluctuations by designing a dynamic threshold function. The hierarchical threshold is dynamically adjusted through scene activity and system stability indicators to achieve adaptive changes in the threshold. During crowded periods, the system automatically increases the abnormality judgment threshold to reduce false alarms caused by normal fluctuations in human flow; when the system load is high, the threshold of low-priority abnormalities is appropriately increased to ensure that system resources are concentrated on processing high-priority events. Deployment tests show that compared with the fixed threshold system, the dynamic threshold mechanism reduces the peak false alarm rate by 57% while maintaining the same detection rate. At the same time, it can maintain a 100% detection rate for key abnormalities when system resources are tight, and the delayed processing time of low-risk abnormalities is only increased by 0.5 seconds. Especially in shopping malls and transportation hubs with large passenger flow fluctuations, this embodiment improves the stability and reliability of the system at different times, reduces the workload of security personnel in handling false alarms, and improves the overall security efficiency.

[0135] like Figure 4 As shown, according to one aspect of the present application, the step of obtaining a cross-camera abnormal event association graph includes:

[0136] S31, based on the abnormal behavior energy classification matrix, applying the preconfigured resource demand prediction model to calculate the predicted resource demand value of each camera node, and generating a global resource demand matrix;

[0137] S32, applying a pre-configured resource allocation optimization model to solve the global resource demand matrix, considering detection priority and system load balancing, and generating a camera resource allocation plan;

[0138] S33. Based on the abnormal behavior energy classification matrix and the camera resource allocation plan, determine the target of attention, build a multi-camera collaborative tracking protocol and combine it with the pre-stored camera topology map for collaborative tracking, and generate a cross-camera abnormal event association map.

[0139] Specifically, a resource allocation optimization model is constructed, and the objective function is: maximize ∑(P i ·D i ) -λ·∑|L i -L avg |; where P i is the detection priority of level i abnormal events, set to [1.0, 0.8, 0.5, 0.3, 0.1]; D i The detection rate of the corresponding abnormal events is estimated by the ratio of resource allocation to required resources: Di = min(1.0, allocated_res / required_res), where allocated_res is the amount of resources allocated to a task or event; required_res is the amount of resources required to complete a task or event; L i is the resource load of processing node i, with a value range of [0, 1]; L avg is the average load of the system; λ=0.3 is the load balancing adjustment coefficient. The Lagrangian relaxation algorithm is used to solve the resource allocation optimization model and obtain the initial resource allocation plan; based on the initial resource allocation plan and combined with the available computing power of the edge nodes, a camera resource allocation plan that considers the network topology is generated.

[0140] This embodiment solves the problems of uneven resource allocation and isolated camera operation in traditional monitoring systems by constructing a resource demand prediction model based on the abnormal behavior energy classification matrix, designing a resource allocation optimization model, and implementing multi-camera collaborative tracking. The new system can dynamically adjust the allocation of computing resources according to the energy level of abnormal events, and achieve full monitoring of high-risk abnormal behaviors through multi-camera collaborative tracking. In an edge computing environment, this embodiment optimizes the efficiency of computing resource utilization. Compared with fixed resource allocation, the overall response time of the system is shortened by 43%, while the continuous tracking rate of abnormal targets is increased from the original 68% to 94%. Especially when the target crosses the blind spot of the camera, the multi-camera collaborative tracking protocol can achieve seamless connection, effectively preventing the loss of high-risk targets, and providing security personnel with a continuous and complete chain of evidence of abnormal behavior.

[0141] According to one aspect of the present application, the step of generating a global resource requirement matrix includes:

[0142] Based on the abnormal behavior energy classification matrix, a resource demand prediction model R(c, t)=base(c)+∑(w i ·Ei )·f(t)+μ·V(c,t);

[0143] Where R(c, t) is the computing resource requirement of camera c at time t, base(c) is the basic resource requirement of camera c, which is determined by the resolution and frame rate. For example, 1080p / 30fps corresponds to base=100 units. i is the total energy value of abnormal events of level i in the surveillance area of ​​the camera, w i The resource allocation weight for level i anomalies is set to [5.0, 3.0, 1.5, 0.8, 0.3]; f(t) is the time decay function, f(t) = exp(-λt), λ=0.05 / min; V(c, t) is the video content complexity of camera c at time t, which is evaluated by calculating the inter-frame difference and the number of moving targets: V(c, t) = 0.7·D_frame +0.3·(N_obj / N_max), D_frame is the normalized frame difference; N_max=50 is the preset value of the maximum number of moving targets in camera monitoring; μ=50 is the complexity adjustment coefficient; N_obj is the number of moving targets detected by camera c at time t;

[0144] Apply the resource demand prediction model to each camera node and calculate the predicted resource demand value;

[0145] Summarize the predicted resource demand values ​​of all cameras to generate a global resource demand matrix.

[0146] This embodiment solves the problem of static resource allocation in traditional systems by constructing a resource demand prediction model. The basic resource demand, abnormal event energy weighting and video content complexity are organically combined to achieve accurate prediction of computing resource demand. Through the time decay function, the system can gradually reduce the resource allocation weight as the duration of the abnormal event increases to avoid long-term resource occupation; through the video content complexity assessment, the system can allocate more resources to complex scenes such as dense crowds or drastic changes in lighting. In practical applications, this embodiment increases the system's processing capacity by 42% while keeping the total amount of resources unchanged. Especially during the peak period of emergencies, the system can still maintain stable operation, and the abnormal event processing delay is reduced from 1.7 seconds of the traditional system to 0.4 seconds, which improves the system's response to sudden security incidents.

[0147] According to one aspect of the present application, the camera resource allocation scheme is also optimized, specifically:

[0148] S321, modeling resource consumption of abnormal events. Based on the abnormal behavior energy classification matrix, a resource consumption model is established for each type of abnormal event; historical data is collected to calculate the average processing resource requirements of abnormal events of different levels; and an abnormal event resource requirement mapping table is output.

[0149] S322, perform time-space related resource demand forecast. Construct time-space related resource forecast function: RT(c, t) = ∑(w i ·E i )·f(t) + ∑∑(w ij ·E j ·K(d ij ))·f(t-τ); the second term represents the impact of abnormal event Ej in adjacent camera j on the resource demand of camera c through spatial correlation function K and time offset τ; w ij is the association weight coefficient between the camera and its adjacent cameras; d ij is the distance between the camera and the adjacent cameras; calculates the predicted resource demand matrix considering event spread and migration, and realizes the prediction of resource demand for possible propagation paths of abnormal events.

[0150] S323. Calculate the adaptive resource allocation priority. Construct a priority scoring function: P(c, t) = θ·RT(c, t)+ (1-θ)·I(c); where I(c) is the business importance index of the camera area, and θ is the balance parameter; calculate the resource allocation priority of each camera node to form a dynamic priority sequence; use the dynamic priority sequence as the key basis for resource allocation.

[0151] S324, solve resource allocation constraint optimization. Construct resource allocation constraint optimization model: maximize ∑(P(c, t)·X(c, t)) constraint: ∑X(c, t)≤R total (total resource constraint); X(c, t) ≥ X min (c) (minimum resource guarantee); |X(c, t) - X(c, t-1)|≤ΔX max (resource fluctuation constraint); X(c, t) is the amount of resources allocated to camera c at time t. The Lagrange multiplier method is used to solve this optimization model and generate an optimized resource allocation solution that satisfies multiple constraints.

[0152] S325, Generate resource scheduling and execution strategy. Based on the optimized resource allocation plan, generate specific camera processing strategies: including key parameters such as sampling rate, processing priority, feature extraction depth, etc.; build a smooth transition mechanism to ensure system stability during dynamic resource adjustment, output the final camera resource scheduling instructions, and guide system resource allocation.

[0153] According to one aspect of the present application, the step of generating a cross-camera abnormal event association graph includes:

[0154] Based on the abnormal behavior energy classification matrix and camera resource allocation plan, determine the targets that need to be monitored;

[0155] Construct a multi-camera collaborative tracking protocol with three core mechanisms: task delegation, resource borrowing, and information sharing;

[0156] For high-energy abnormal behaviors, start collaborative tracking of adjacent cameras and generate collaborative tracking tasks;

[0157] Build a camera topology map based on the spatial relationship and coverage of cameras for collaborative task distribution;

[0158] By performing collaborative tracking tasks, a cross-camera abnormal event correlation graph containing spatiotemporal correlation information is generated.

[0159] This embodiment solves the problem of target tracking interruption caused by the independent operation of cameras in traditional surveillance systems by determining the target of interest, designing a multi-camera collaborative tracking protocol, and constructing a camera topology map. It enables adjacent cameras to work together to achieve continuous tracking of high-energy abnormal behaviors. The task delegation mechanism in the collaborative tracking protocol allows the main camera to assign tracking tasks to adjacent cameras; the resource borrowing mechanism enables resource-constrained cameras to temporarily obtain additional computing power; and the information sharing mechanism ensures that target features are seamlessly transmitted between different cameras. In the actual deployment test of a large shopping mall, this embodiment increased the target cross-camera continuous tracking rate from the original 64% to 91%, and reduced the average target loss time from 4.6 seconds to 0.8 seconds, and can maintain stable tracking of abnormal targets even in crowded areas. This enhances the monitoring system's ability to monitor suspicious persons throughout the entire process and provides a complete chain of evidence for post-event evidence collection of security incidents.

[0160] According to one aspect of the present application, the step of generating optimized configuration parameters and feeding back to the resource optimization step includes:

[0161] S41, applying spatiotemporal consistency verification to the cross-camera abnormal event association graph, generating verified abnormal events, and generating abnormal event impact range graphs and graded warning information based on their energy levels and spatial distributions;

[0162] S42, collecting system operation data to form a system performance indicator set, applying a pre-configured resource benefit evaluation model to calculate resource benefit indicators, optimizing resource allocation weight parameters, and generating optimized configuration parameters;

[0163] S43, feeding back the optimized configuration parameters to the resource demand prediction model and the resource allocation optimization model, and updating the pre-configured typical abnormal pattern library and feature extraction configuration parameters at the same time, to perform closed-loop optimization.

[0164] Specifically, spatiotemporal consistency verification is applied to the cross-camera abnormal event correlation graph to eliminate isolated false alarms and obtain verified abnormal events; based on the energy level and spatial distribution of the verified abnormal events, a radial diffusion algorithm is applied to generate an abnormal event impact range map; based on the type, level, impact range and urgency of the abnormal event, graded warning information containing processing suggestions is generated; for different levels of warnings, an adaptive warning information transmission path is constructed to ensure that key warnings can reach response personnel in a timely manner.

[0165] Feedback the optimized configuration parameters to the resource demand prediction model and resource allocation optimization model to update the model parameters; extract features of new abnormal patterns detected during system operation and update the typical abnormal pattern library; based on operation data, update the scenario classification and feature extraction strategies, and optimize the feature extraction configuration parameters; build an exception handling knowledge base, record abnormal event handling experience, and form a closed-loop optimization mechanism.

[0166] This embodiment realizes closed-loop optimization and adaptive adjustment of the system by verifying the spatiotemporal consistency of the cross-camera abnormal event correlation graph, monitoring system performance indicators, and evaluating resource efficiency. It solves the limitation of the traditional early warning system of "only reporting but not optimizing", and establishes a virtuous cycle of resource allocation, early warning generation, and system optimization. The system continuously optimizes resource allocation parameters and abnormal energy calculation weights by collecting performance indicators including abnormal detection rate, resource utilization, response time, and false alarm rate, so that the system performance continues to improve with the running time. The actual deployment test shows that this embodiment improves the resource efficiency index of the system by 29% and the abnormal detection accuracy by 17% after 3 months of operation. At the same time, the abnormal detection capability of the system per unit power consumption is improved by 32%, which reduces the long-term operation cost and improves the sustainability of the system.

[0167] According to one aspect of the present application, the step of generating optimized configuration parameters includes:

[0168] Collect system operation data including abnormal event detection rate, resource utilization, response time, and false alarm rate at all levels to form a set of system performance indicators;

[0169] Construct a resource benefit evaluation model E=α·DR+β·(1 / RT)+γ·RU-δ·FP, where E is the resource benefit value, DR is the anomaly detection rate, RT is the system response time, RU is the resource utilization rate, FP is the false alarm rate, and α, β, γ, and δ are adjustable weight coefficients;

[0170] Applying the resource benefit evaluation model to the system performance indicator set to calculate the resource benefit indicator;

[0171] Based on the resource efficiency index, the stochastic gradient descent algorithm is applied to optimize the resource allocation weight parameters and generate the optimized configuration parameters.

[0172] This embodiment solves the problem of the lack of quantitative standards for resource allocation benefits in traditional systems by designing a resource benefit evaluation model. The model comprehensively considers four key indicators: abnormal detection rate (DR), system response time (RT), resource utilization (RU) and false alarm rate (FP), and creates a unified resource benefit evaluation system. Based on this model, the system can apply the stochastic gradient descent algorithm to continuously optimize the resource allocation weight parameters to form a self-optimization cycle. In the deployment test of the smart factory, this embodiment enables the system to increase the abnormal detection rate by 18%, reduce the average response time by 32%, reduce the false alarm rate by 23%, and increase the resource utilization by 26% under the same resource conditions after two months of operation. In particular, in response to changes in resource demand in different time periods, the system can automatically adjust the parameter weights, such as increasing the response time weight during the peak period of factory shift changes and increasing the detection rate weight during the low peak period at night, realizing the time intelligence of resource allocation.

[0173] In a specific embodiment of the present application, a method for identifying and warning abnormal behavior based on video image processing is applicable to video surveillance systems in public places such as shopping malls, airports, and subway stations. The system consists of 32 high-definition network cameras (1080p resolution, 30fps frame rate) distributed in different areas to form a distributed monitoring network. Video processing is performed through edge computing devices, and the central server is responsible for system coordination and warning generation. It should be noted that the process of introducing the activity information data source and fusing it with the video image has been described in the above embodiment and will not be described in detail here.

[0174] The specific steps are as follows:

[0175] Step 1: First, pre-process the raw video stream data collected from the distributed monitoring network. Taking the video of a monitoring point in a shopping mall as an example, the system performs the following steps:

[0176] Timestamp synchronization processing: synchronize the timestamps of each camera to eliminate the time deviation between devices. Calculate the time deviation △t = t_local - t_reference, where t_local is the local camera timestamp and t_reference is the reference timestamp. Correct the video frame timestamp: t_corrected = t_original - △t, where t_original is the original video frame timestamp and t_corrected is the corrected video frame timestamp. Perform adaptive noise reduction processing: automatically select the most suitable noise reduction algorithm according to the scene lighting conditions, apply a Gaussian filter with a parameter σ=1.2 in areas with sufficient light (brightness value L>180); apply a bilateral filter with a parameter σ in low-light areas (brightness value L<80). d =3.0,σ r = 0.1. Perform resolution unification processing: adjust the videos of cameras with different specifications to a unified 1280×720 resolution and 25fps frame rate to generate a standardized video frame sequence F = {f1, f2, ..., fn}.

[0177] Perform scene analysis and classification: Extract GIST descriptor G = [g1, g2, ..., g512] from the standardized video frame sequence, where each element gi represents a global feature component of the scene. Apply density-weighted K-means algorithm to automatically classify the scene: Initialize K cluster centers C = {c1, c2, ..., cK}, K = 5; Calculate the distance d(G, ci) = ||G - ci|| from each sample point to each cluster center 2 ; Calculate the local density ρ(G) = Σexp(-||G - Gj|| 2 / 2σ 2 ), j∈the set of neighboring points; assign the sample G to the nearest cluster center, but give a greater weight w to the high-density area G，i = 1 / (d(G, ci) × (1 + λ·ρ(G))), λ=0.3, select the maximum weight w G，i Corresponding cluster i; update the cluster center, return to calculating the distance from each sample point to each cluster center until convergence, and obtain the scene type label S∈{"open area", "entrance passage", "elevator area", "cash register area", "rest area"}.

[0178] Select the feature extraction strategy. According to the scene type label S, select the optimal feature extraction method combination from the feature extraction strategy library. For example, for the "open area" scene, select the configuration parameter P_open = {optical flow feature weight: 0.5, trajectory feature weight: 0.3, posture feature weight: 0.2}; for the "cashier area" scene, select the configuration parameter P_cashier = {trajectory feature weight: 0.2, posture feature weight: 0.5, interaction feature weight: 0.3}.

[0179] Perform multimodal feature extraction and fusion. Extract corresponding features according to the configuration parameters. Take the "open area" as an example: extract optical flow features OF = {of1, of2, ..., ofm}, where ofi represents the average optical flow vector of the i-th area; extract trajectory features TR = {tr1, tr2, ..., trp}, where tri represents the trajectory point set of the i-th detection target; extract posture features PS = {ps1, ps2, ..., psq}, where psi represents the key point coordinate set of the i-th human body.

[0180] Apply sparse coding algorithm to fuse multimodal features: construct feature dictionary D = [D_OF | D_TR | D_PS]; perform sparse representation of each feature vector min ||α||1 subject to ||F - Dα||2 ≤ ε; weighted fusion sparse coefficient α_fused = w_OF·α_OF + w_TR·α_TR + w_PS·α_PS. Where D_OF is the feature dictionary of optical flow features; D_TR is the feature dictionary of trajectory features; D_PS is the feature dictionary of posture features; α is the sparse coefficient vector; ε is the tolerance of reconstruction error; w_OF, w_TR, w_PS are the weighted coefficients of optical flow, trajectory, and posture features respectively; α_OF, α_TR, α_PS are the sparse coefficients of optical flow, trajectory, and posture features respectively.

[0181] Analyze the temporal pattern. Use a dynamic time window H(t) = max(H_base, β·V(t)) to analyze the temporal changes of features, where H_base = 25 frames is the base window length, V(t) is the current scene activity, and β = 0.8 is the adjustment coefficient. Calculate the temporal statistics of the feature vector in the window, including the mean μ_feat, standard deviation σ_feat, and change rate δ_feat, and generate a scene-adaptive behavior feature vector B = [α_fused, μ_feat, σ_feat, δ_feat].

[0182] Step 2: Based on the scene-adaptive behavior feature vector B, perform abnormal behavior energy assessment. Construct a dynamic background model: Use the Gaussian mixture model (GMM) to construct a dynamic background model M. Each scene type maintains K Gaussian components (K=5), and the parameters of the i-th Gaussian component are {μi, Σi, ωi}, which represent the mean, covariance matrix and weight. For the newly observed feature vector B, update the parameters of each Gaussian component: if ||B - μi|| Σi < 2.5σ (matching the i-th component): ωi_new = (1-η)·ωi + η, η=0.01 is the learning rate; μi_new = (1-ρ)·μi + ρ·B, ρ=η / ωi; Σi_new = (1-ρ)·Σi + ρ·(B-μi)(B-μi) T ; else: replace the component with the smallest weight and initialize it to {B, Σinit, ωmin}. T represents transpose; Σinit is the initial value of the covariance matrix of the newly initialized Gaussian component; ωmin is the initial weight value of the new component, which is usually set to a very small value.

[0183] Define the normal behavior boundary: Use a single-class support vector machine to construct the normal behavior boundary. Select the radial basis kernel function K(B, B') = exp(-γ||B-B'|| 2 ), parameter γ=0.1, the normal behavior boundary M is expressed as: f(B) = Σαi·K(B,Bi) – ρ; where Bi is the support vector set, B' is a vector in the support vector set, αi is the Lagrange multiplier, and ρ is the bias parameter. If f(B)<0, then B is within the normal behavior boundary.

[0184] Calculate the multidimensional anomaly energy function: For the current behavior feature vector B, construct the anomaly energy function: E(B) = α·Ds(B, M) + β·Dt(B, H) + γ·C(B) +δ·I(B). Where Ds(B, M) = max(0, -f(B)) / σM, σM is the boundary scale parameter; Dt(B, H) = min Bi∈H ||BB i || 2 · exp(-λ·(tt i )), λ=0.05 is the time decay factor; C(B) is calculated as a combination of trajectory complexity and speed change frequency: C(B) = 0.6·Cc(TR) + 0.4·Cv(OF), where Cc is the trajectory curvature and Cv is the speed change frequency; I(B) is related to the degree to which the monitored object approaches the sensitive area: I(B) = Σw i exp(-d i2 / 2σ i 2 ), d i is the distance to the i-th sensitive area, w i is the weight of the region; the initial values ​​of α, β, γ, and δ are set to α=0.4, β=0.3, γ=0.2, and δ=0.1.

[0185] Calculate the local energy density: Construct the local energy density function ρ(B, r): ρ(B, r) = ΣE(B j )·w(d(B,B j )), j∈{j | d(B, B j ) ≤ r}; where B j is the other behavior within the radius r, d(B, B j ) is the Euclidean distance in the feature space, and w is the Gaussian kernel function weight: w(d) = exp(-d 2 / 2σ 2 ), σ = r / 3 is the influence range parameter, and r = 0.2 is the search radius in the standardized feature space.

[0186] Analyze time series energy fluctuations: Construct energy fluctuation function Φ(B, t) to analyze the energy variation pattern of the behavior within the time window: Φ(B, t) = [Δe1, Δe2, ..., Δe 10 ,σ,τ]; where Δe i Indicates the rate of change of energy at consecutive time points: Δe i = (E(B, t i ) - E(B, t i₋1 )) / E(B,t i₋1 ), σ is the standard deviation of the energy change sequence, and τ is the autocorrelation coefficient (lag=2). According to the values ​​of σ and τ, anomaly classification is performed.

[0187] Calculate the fusion-enhanced energy: Design the fusion-enhanced energy function: E'(B) = E(B)·[1 + λ1·f(ρ(B, r)) + λ2·g(Φ(B, t))].

[0188] Perform dynamic threshold calculation and classification: Design dynamic threshold function for different levels of abnormalities: θ i (t) = θ i，base ·[1+μ·V(t)-ν·S(t)] According to the enhanced abnormal energy value E'(B) and the dynamic threshold θ i (t) is compared to classify abnormal behaviors. Finally, the abnormal behavior energy classification matrix EM is generated, which contains the abnormal target ID, location, abnormal energy value and classification information detected in each camera area.

[0189] Step 3: Based on the abnormal behavior energy classification matrix EM, perform resource optimization and collaborative tracking. Build a resource demand prediction model to calculate the predicted resource demand value of each camera node c at time t: R(c, t) = base(c) + ∑(w i ·E i )·f(t) + μ·V(c,t).

[0190] Apply the resource allocation optimization model to maximize the objective function under the constraints: maximize ∑(P i ·D i ) -λ·∑|L j -L avg |. Apply the Lagrangian relaxation algorithm to solve this optimization problem: construct the Lagrangian function L(x, λ); iteratively solve the dual problem and update the Lagrangian multiplier; optimize the original problem variables using the projected gradient method; check the convergence conditions, return the iterative solution to the dual problem or output the results; generate the camera resource allocation plan RA, which includes the resource allocation amount, processing priority and sampling rate settings for each camera.

[0191] Perform multi-camera collaborative tracking: Based on the abnormal behavior energy classification matrix EM and the resource allocation scheme RA, determine the target set T = {t1, t2, ..., t n}, each target contains ID, location, movement direction and abnormal energy level. Constructing a multi-camera collaborative tracking protocol includes three core mechanisms: Task delegation mechanism: When the target t i When the camera c1 is about to leave its field of view, c1 sends a task delegation request to the adjacent camera c2 that may receive the target: TaskAssign(t i, feature_vector, predicted_exit_time, predicted_exit_point); TaskAssign is a task assignment request, feature_vector is the feature vector of the target, predicted_exit_time is the time when the target is predicted to leave the current camera's field of view, and predicted_exit_point is the position where the target is predicted to leave the field of view; Resource borrowing mechanism: When the camera resources are insufficient, a resource borrowing request is sent to the adjacent low-load camera: ResourceBorrow(amount, duration, priority); ResourceBorrow is a resource borrowing request, mount is the amount of resources that need to be borrowed, duration is the duration of the borrowed resources, and priority is the priority of the borrowing request; Information sharing mechanism: cameras regularly share abnormal target information, and the frequency is dynamically adjusted according to the target abnormality level: InfoShare(target_list, anomaly_levels, interval=5s / level), where InfoShare is the information sharing mechanism, target_list is the target set that needs to be shared, anomaly_levels is the abnormality level of the corresponding target, and interval=5s / level means that the information sharing frequency is dynamically adjusted according to the abnormality level. For example, the higher the abnormality level, the faster the frequency of information sharing (such as sharing once every 5 seconds). Construct a camera topology graph G = (V, E) based on the spatial relationship and coverage of the camera, where V is a set of camera nodes, each of which contains location, coverage and resource capacity information; E is a set of edges, indicating the overlap or adjacency between the cameras, and each edge contains the size of the overlapping area and transmission delay information.

[0192] For targets with abnormality levels reaching "high risk" or above, the collaborative tracking process is started: the main camera c_main detects the high-risk abnormal target t; the possible moving path of the target is predicted P = {p1, p2, ..., p k} and probability; According to the path P and the camera topology map G, determine the camera set C_next that may receive the target; The main camera sends a task delegation request to the camera set C_next; The cameras in the camera set C_next adjust the processing parameters and prepare to receive the target; After the target enters the field of view of the new camera, perform feature matching confirmation: match_score = feature_match(t_feature, candidate_feature); if match_score > threshold_match(0.7): Confirm that it is the same target and update the global target state. Generate a cross-camera abnormal event association graph EG, which contains the movement trajectory, timestamp and abnormal energy change information of the abnormal target between different cameras. Among them, match_score is the score of feature matching, feature_match is the feature matching function, t_feature is the feature vector of the current target, candidate_feature is the feature vector of the candidate target, and threshold_match is the matching threshold.

[0193] Step 4: Based on the cross-camera abnormal event association graph EG, the system generates warnings and optimizes strategies. Perform spatiotemporal consistency verification: Apply spatiotemporal consistency verification to the cross-camera abnormal event association graph to eliminate false alarms: for each abnormal event e in EG: if event duration < min_duration (level): mark e as "pending verification"; if the event is detected in multiple cameras and is consistent in time and space: mark e as "confirmed"; if the event energy fluctuates significantly in a short period of time: perform additional verification steps; generate a verified abnormal event set VE. Calculate the impact range of abnormal events: Based on the energy level and spatial distribution of the verified abnormal events, apply the radial diffusion algorithm to generate the impact range map: for each confirmed abnormal event e in VE: Initialize the impact area A = the location of the event; Determine the maximum diffusion radius R = base_radius × level_multiplier[e.level] according to the event energy level E; Where base_radius is the basic diffusion radius, level_multiplier is the multiplier factor related to the energy level of the abnormal event, and e.level is the energy level of the abnormal event e; Apply the Fast Marching Method to calculate the impact propagation velocity field; Generate contour lines to represent the impact range boundary; Generate the abnormal event impact range map IR. Generate graded warning information based on the type, level, impact range and urgency of the abnormal event: for each confirmed abnormal event e in VE: Select the warning template based on e.type; Fill in key information: location, time, abnormal description, impact range, threat level; Determine the warning color and urgency according to e.level; Generate processing suggestions and response guidelines; Generate a graded warning information set WI, which is displayed and pushed to relevant personnel through the system control center. Where e.type is the type of abnormal event e.

[0194] Conduct resource efficiency evaluation: collect system operation data to form a system performance indicator set PI, including: anomaly detection rate DR, the proportion of abnormal events of each level that are successfully detected; system response time RT, the average delay from the occurrence of anomalies to the generation of warnings; resource utilization rate RU, the effective use ratio of allocated resources; false alarm rate FP, the proportion of events marked as abnormal by the system but actually normal. Construct a resource efficiency evaluation model: E = α·DR + β·(1 / RT) + γ·RU -δ·FP; where α=0.4, β=0.3, γ=0.2, δ=0.1 are adjustable weight coefficients. Based on the resource efficiency indicator E, apply the stochastic gradient descent algorithm to optimize the resource allocation weight parameters: Initialize the weight parameters w = [w1, w2, ..., w n]; Set learning rate η = 0.01, decay factor λ = 0.995; for iterations i = 1 to max_iter: Calculate the current benefit value E and gradient ▽E; w_new = w- η·▽E; Apply constraints to ensure that the new weight parameter w_new satisfies the valid range if |E_new - E| < ε: Break; w = w_new; η = η × λ; Generate optimization configuration parameters OP, including resource allocation weights, abnormal energy function weights, and threshold parameters. Where max_iter is the maximum number of iterations of the algorithm, ε is the threshold of the convergence condition, and E_new is the new benefit value calculated in the current iteration.

[0195] Perform strategy adaptive update. Feedback the optimized configuration parameters OP to each module of the system: Update the weight parameter w of the resource demand prediction model i ; Update the priority parameter P of the resource allocation optimization model i ; Update the weight parameters α, β, γ, δ of the anomaly energy function; Update the configuration parameters of the feature extraction strategy; Build an anomaly handling knowledge base KB to record the anomaly event handling experience and optimization history to form a closed-loop optimization mechanism.

[0196] This embodiment was deployed and tested in a large shopping mall for 6 months. The overall performance of the system is as follows: the comprehensive accuracy rate reached 92.7%, which is 31.5% higher than the traditional single-dimensional evaluation system. The recognition rates of different types of abnormalities are as follows: abnormal lingering behavior: 95.3%; abnormal rapid movement: 91.8%; abnormal wandering back and forth: 89.5%; abnormal group gathering: 94.1%. The overall false alarm rate of the system is 7.3%, which is 67.8% lower than the traditional fixed threshold system. The false alarm rate is controlled at 12.5% ​​during peak traffic hours and drops to 4.2% during non-peak hours. The average resource utilization rate of the system is increased by 38.2%, and the computing resource consumption is reduced by 41.5% when processing the same number of camera video streams. The average time for abnormal behavior to be detected and generate an early warning is 0.8 seconds, which is 2.7 seconds earlier than the traditional system to issue an early warning, which buys valuable time for security personnel to respond. The cross-camera target tracking success rate reached 91.3%, and the average target loss time was reduced from the original 4.3 seconds to 0.7 seconds, effectively realizing the full monitoring of high-risk targets. After the system was put into operation for 3 months, the resource efficiency index increased by 29.4%, and the anomaly detection accuracy increased by 16.8%, achieving continuous improvement in performance. It can be seen from the above embodiments that the abnormal behavior recognition and early warning system based on this embodiment can achieve high-accuracy abnormal behavior detection and early warning in complex public places, thereby improving the practicality and reliability of the video surveillance system.

[0197] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for identifying and warning abnormal behavior based on video image processing, characterized in that: The following steps are involved: Collect basic data and perform scene-adaptive feature extraction based on it to generate scene-adaptive behavior feature vectors; basic data includes original video stream data, user setting data and activity information data; Perform multi-dimensional abnormal behavior energy evaluation and classification on the scene-adaptive behavior feature vector to obtain an abnormal behavior energy classification matrix; Based on the abnormal behavior energy classification matrix, dynamic resource optimization and multi-camera collaborative tracking are performed to obtain a cross-camera abnormal event correlation graph; Verify the cross-camera abnormal event correlation graph and generate graded warning information.

2. The method according to claim 1, characterized in that The steps of performing multi-dimensional abnormal behavior energy assessment and classification and obtaining an abnormal behavior energy classification matrix include: Based on the historical data of scene-adaptive behavior feature vectors, a time-sensitive background model with memory decay characteristics is constructed, normal behavior boundaries are established, and a typical abnormal pattern library is extracted; Based on the typical abnormal pattern library, an abnormal energy function is constructed; for the adaptive behavior feature vector of the current scenario, the abnormal energy function is applied to calculate the abnormal energy value including spatial distance, temporal distance, complexity coefficient and business impact factor, and generate abnormal behavior indicators; Establish a regional value map, conduct a multi-dimensional comprehensive assessment based on abnormal behavior indicators, classify abnormal behaviors, and generate an abnormal behavior energy classification matrix.

3. The method according to claim 2, characterized in that The abnormal energy function is applied to calculate the abnormal energy value including the spatial distance, the temporal distance, the complexity coefficient and the business impact factor. The steps of generating the abnormal behavior index include: Construct the abnormal energy function E(b) = α·Ds(b, M) + β·Dt(b, H) + γ·C(b) + δ·I(b); Where Ds(b, M) represents the spatial distance between behavior b and the normal behavior boundary M, Dt(b, H) represents the temporal distance between behavior b and the historical behavior sequence H, C(b) represents the complexity coefficient of behavior b, I(b) represents the business impact factor of behavior b, and α, β, γ, and δ are adjustable weight parameters; Applying the abnormal energy function to the current scene adaptive behavior feature vector to calculate the abnormal energy value; Based on the abnormal energy value and the preset judgment criteria, the degree of abnormal behavior is determined and an abnormal behavior indicator is generated.

4. The method according to claim 2, characterized in that: The steps of establishing a regional value map, combining abnormal behavior indicators for multi-dimensional comprehensive evaluation, and grading abnormal behaviors to generate an abnormal behavior energy grading matrix include: Based on the business importance of the preset monitoring areas, a regional value map is constructed to show the relative value of each area; The abnormal behavior index and the regional value map are weighted and combined to construct a weighted abnormal importance index that reflects the importance of abnormal behavior; Applying the hierarchical analysis algorithm, a multi-dimensional comprehensive evaluation of the weighted abnormal importance index is performed according to the pre-stored abnormal behavior type, occurrence location and impact range to obtain the evaluation result; Based on the evaluation results, abnormal behaviors are divided into five levels: critical, high-risk, medium-risk, low-risk, and prompt, and an abnormal behavior energy classification matrix is ​​generated.

5. The method according to claim 3, characterized in that: The weight parameters α, β, γ, and δ in the abnormal energy function are dynamically adjusted through the following steps: Collect historical data containing confirmed abnormal behaviors and extract historical anomaly detection results; Based on the historical anomaly detection results, the particle swarm optimization algorithm is applied to calculate the weight parameter combination that maximizes the detection accuracy; Update the weight parameters α, β, γ, δ in the anomaly energy function, maintaining the constraint condition of α+β+γ+δ=1; The weight parameters are adjusted periodically according to the current scene changes to make the abnormal energy function adapt to the characteristics of different scenes.

6. The method according to claim 3, characterized in that After calculating the abnormal energy value, the step of calculating the local energy density is also included: Construct a local energy density function ρ(b, r) to represent the abnormal energy density within a radius r around behavior b; ρ(b, r) = ∑E(bi) w(d(b, bi)), where bi is other behaviors within the radius r, E(bi) is its abnormal energy value, d(b, bi) is the distance between behaviors b and bi, and w is a weight function based on distance; Gaussian kernel function is used as the weight function w(d)=exp(-d 2 / 2σ 2 ), where σ is an adjustable influence range parameter and d is the distance; The local energy density values ​​of behavior b are calculated in aggregate to capture the local aggregation effect of abnormal behaviors.

7. The method according to claim 6, characterized in that After calculating the abnormal energy value, the following steps are also included: constructing an energy fluctuation function to analyze the time series energy feature vector: Construct the energy fluctuation function Φ(b, t)=[Δe1, Δe2, ..., Δe n , σ, τ], which is used to analyze the energy variation pattern of behavior b in the time window t; where n is the number of discrete time points contained in the time window t; Calculate the rate of energy change Δe at consecutive time points i =(E(b, t i )-E(b,t i₋1 )) / E(b,t i₋1 ), generate an energy change sequence; where i is the index of the time point; Calculate the fluctuation characteristics of the energy change sequence, including the standard deviation σ and the autocorrelation coefficient τ, to distinguish stable, fluctuating and mutation anomalies; Combining the energy change sequence and fluctuation characteristics, a time series energy feature vector is generated to characterize the time series development characteristics of abnormal behavior.

8. The method according to claim 7, characterized in that The step of constructing a fusion-enhanced energy function is further included: Construct the fusion enhanced energy function E'(b)=E(b)·[1+λ1·f(ρ(b, r))+λ2·g(Φ(b, t))], where E(b) is the original abnormal energy value, ρ(b, r) is the local energy density value, Φ(b, t) is the time series energy feature vector, λ1 and λ2 are adaptive weight coefficients; f and g are nonlinear mapping functions; The fusion enhanced energy function is applied to calculate the enhanced anomaly energy index to achieve differentiated evaluation of isolated anomalies, clustered anomalies and evolving anomalies.

9. The method according to claim 1, characterized in that: The steps for performing dynamic resource optimization and multi-camera collaborative tracking to obtain a cross-camera abnormal event correlation graph include: Based on the abnormal behavior energy classification matrix, the pre-configured resource demand prediction model is applied to calculate the predicted resource demand value of each camera node and generate a global resource demand matrix; Apply the pre-configured resource allocation optimization model to solve the global resource demand matrix, consider detection priority and system load balancing, and generate a camera resource allocation plan; Based on the abnormal behavior energy classification matrix and camera resource allocation scheme, the target of attention is determined, a multi-camera collaborative tracking protocol is constructed and combined with the pre-stored camera topology map for collaborative tracking, and a cross-camera abnormal event correlation map is generated.

10. The method according to claim 9, characterized in that The steps of verifying the cross-camera abnormal event correlation graph and generating graded warning information, evaluating resource allocation efficiency, generating optimized configuration parameters and feeding back to the resource optimization step include: Apply spatiotemporal consistency verification to the cross-camera abnormal event association graph to generate verified abnormal events, and generate abnormal event impact range graphs and graded warning information based on their energy levels and spatial distribution; Collect system operation data to form a system performance indicator set, apply the pre-configured resource benefit evaluation model to calculate the resource benefit indicator, optimize the resource allocation weight parameters, and generate the optimized configuration parameters; The optimized configuration parameters are fed back to the resource demand prediction model and the resource allocation optimization model, and the pre-configured typical abnormal pattern library and feature extraction configuration parameters are updated to perform closed-loop optimization.

Images