An abnormal event identification method and system of an intelligent security camera

Abstract

The application provides an abnormal event identification method and system of an intelligent security camera, and is applied to the field of data processing; the application can not depend on a fixed abnormal event category set in advance, dynamically analyze an abnormal evolution process of a target object based on continuous behavior changes, spatial activity rules and interaction relationships among objects in a monitoring area, can discover unknown abnormal behaviors that cannot be covered by a traditional abnormal classification model, can construct an abnormal behavior evolution path, extract regular characteristics corresponding to undefined abnormal events, and combine historical abnormal event data to perform clustering analysis, can realize automatic induction of similar abnormal trends and generation of abnormal labels, and make the system have the ability of continuously expanding an abnormal identification range.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a method and system for identifying abnormal events in an intelligent security camera. Background Technology

[0002] With the development of applications such as smart communities, industrial parks, and unmanned locations, intelligent security cameras are gradually transforming from traditional video acquisition devices into video sensing terminals with intelligent analysis capabilities. Currently, intelligent security cameras typically use target detection algorithms, behavior recognition models, and event classification models to analyze personnel activities, vehicle status, and environmental changes within the monitored area, enabling automatic identification of abnormal events such as intrusions, loitering, falls, and gatherings.

[0003] However, existing smart cameras primarily focus on whether an event matches a specific anomaly category, lacking the ability to continuously learn and analyze deviations from normal behavioral patterns in monitored scenarios. Therefore, they fail to detect a large number of undefined, novel anomaly events. For example, in industrial park monitoring scenarios, behaviors such as people lingering near equipment for extended periods, repeatedly moving between areas, or objects being moved abnormally may not fall under traditional definitions of falls, intrusions, or fights, but these behaviors could indicate equipment malfunctions, human error, or security risks. Summary of the Invention

[0004] This invention aims to solve the problem of how to enable intelligent security cameras to break free from their dependence on preset abnormal event categories. By analyzing the behavior of targets, spatial activity patterns, and event evolution processes within the monitored area, it enables the proactive discovery and identification of unknown abnormal events, providing an abnormal event identification method and system for intelligent security cameras.

[0005] The present invention employs the following technical means to solve the technical problem: This invention provides a method for identifying abnormal events in intelligent security cameras, comprising: Based on video data pre-collected by the camera in the target monitoring area, the object type in the target monitoring area is identified. According to the target behavior feature parameters of the object type in a continuous time period, corresponding target behavior state data is generated. The target behavior state data is matched and analyzed with a preset set of behavior benchmark features to calculate the behavior deviation feature parameters in the current monitoring scene. The object type specifically includes people, vehicles and objects. The target behavior feature parameters specifically include position change, movement speed, dwell time, activity frequency and interaction relationship. Determine whether the behavioral deviation feature parameter exceeds a preset abnormal deviation threshold; If the time exceeds the limit, the video data is split into behavioral states in the continuous video segments according to the time order within the continuous time period to obtain the abnormal stage state of the target object. Based on the abnormal stage state, the target behavior evolution sequence of the target object is constructed, the transition relationship between the abnormal stage states is collected, and the corresponding abnormal behavior evolution path is obtained. The abnormal stage state specifically includes the state before the abnormal behavior occurs, the abnormal change process state, and the current abnormal state. Determine whether the evolution process of the abnormal behavior evolution path matches the preset abnormal event characteristics; If there is no match, abnormal candidate objects are marked from the abnormal behavior evolution path, the regular characteristics of the abnormal candidate objects are extracted, corresponding abnormal event description features are generated, and cluster analysis is performed between the abnormal event description features and historical abnormal event data to identify undefined abnormal behaviors with similar evolution trends. The undefined abnormal behaviors are divided into preset abnormal event clusters, and corresponding abnormal labels are generated for the abnormal event clusters. Through the abnormal labels, the abnormal event identification range of the monitoring environment is dynamically expanded. The regular characteristics specifically include behavioral change patterns, spatial activity patterns, and target interaction patterns.

[0006] Furthermore, after the step of splitting the continuous video segments of the video data into behavioral states according to the time sequence within a continuous time period to obtain the abnormal stage state of the target object, the method further includes: Based on anomaly deviation analysis, risk video clips are continuously correlated with target objects in the risk video clips over time. The change information of the target objects at different time nodes is fused to generate a continuous behavioral change trajectory of the target objects. The change information specifically includes changes in position, changes in movement direction, changes in behavior duration, and changes in interaction with surrounding objects. Determine whether the continuous behavior change trajectory meets the preset continuous evolution conditions; If so, based on the correlation between the states of the abnormal stages, identify the changing nodes in different abnormal stages, select the time nodes with the fastest increase in abnormality from the changing nodes, dynamically classify the time nodes as inflection points of abnormal event evolution, and extract behavioral combination features that are not covered by existing abnormal event categories based on the inflection points of abnormal event evolution.

[0007] Furthermore, the step of constructing the target behavior evolution sequence of the target object based on the abnormal stage states, collecting the transition relationships between the abnormal stage states, and obtaining the corresponding abnormal behavior evolution path further includes: Based on the abnormal stage state, the target object state information corresponding to each abnormal stage is obtained. According to the time sequence relationship between different abnormal stages, the target object state information is associated and matched to generate corresponding stage association information. The target object state information specifically includes the spatial area where the target object is located, the duration of the behavior, the changes in the activity range, and the changes in the environment. Determine whether adjacent abnormal stages in the stage association information meet the preset continuous change condition; If so, then the state evolution features between adjacent abnormal stages are collected, the stage transition features of the target object in the process of behavioral change are extracted, and based on the stage transition features, the behavioral evolution features of the target object between different abnormal stages are identified. The key change nodes when the abnormal state evolves from the initial stage to the subsequent stage are divided. Through the key change nodes, the state association structure between abnormal stages is established, and abnormal stages with continuous evolution relationship are combined to generate the corresponding continuous abnormal state chain. The state evolution features specifically include the state change direction, change magnitude and change trend, and the behavioral evolution features specifically include behavior enhancement, behavior migration and behavior reversal.

[0008] Furthermore, the step of marking abnormal candidate objects from the abnormal behavior evolution path, extracting the regular characteristics of the abnormal candidate objects, and generating corresponding abnormal event description features further includes: Based on the abnormal behavior evolution path, the change information set corresponding to each abnormal stage is obtained, the correlation analysis of different state nodes in the abnormal behavior evolution path is performed, the change relationship between each state node is constructed, and the target objects with continuous change characteristics are screened according to the change relationship to generate the corresponding candidate object set. The change information set specifically includes target object state change information, spatial activity change information, and related object change information. Determine whether the target object in the candidate object set satisfies the preset abnormal evolution association condition; If so, then obtain the behavioral performance information of the target object in different evolutionary stages, collect the change patterns of the target object in different dimensions based on the behavioral performance information, construct the behavioral description structure of the target object based on the correlation between the change patterns, and divide the change combination features of existing abnormal categories through the behavioral description structure. The different dimensions specifically include time dimension, space dimension and interaction dimension.

[0009] Furthermore, the step of determining whether the behavioral deviation feature parameter exceeds a preset abnormal deviation threshold also includes: Based on the aforementioned behavioral deviation feature parameters, a corresponding deviation association structure is established; Determine whether the deviation factors in the deviation association structure meet the preset association evolution conditions; If so, based on the deviation association structure, the main influencing factors in the target object's behavior change process are obtained, the evolution order among the main influencing factors is collected, and based on the evolution order, the abnormal changes generated by the target object in a continuous time period are divided into stages to identify the deviation process caused by behavior change. Specifically, the deviation process includes the local deviation process caused by a single behavior change and the composite deviation process formed by the combined action of multiple behavior factors.

[0010] Furthermore, the step of determining whether the evolution process of the abnormal behavior evolution path can match the preset abnormal event characteristics also includes: Based on the abnormal behavior evolution path, the correlation strength between each evolution stage is obtained. According to the correlation strength, the evolution structure in the abnormal behavior evolution path is analyzed to generate path structure parameters of the abnormal behavior development law. Determine whether the path structure parameters meet the preset path stability conditions; If so, based on the stage association relationship in the abnormal behavior evolution path, the evolution nodes in the process of the abnormal behavior of the target object changing from the initial state to the subsequent state are extracted, the change connection relationship between the evolution nodes is identified, the development links in the abnormal behavior development process are determined, and the abnormal behavior evolution logic chain is constructed. The development links specifically include the behavior triggering link, the state transition link, and the risk diffusion link.

[0011] Furthermore, the step of identifying the object type within the target monitoring area based on pre-collected video data from the camera also includes: The basic information of objects within the target monitoring area is obtained, and the corresponding object recognition result is generated based on the positional change relationship of the target object in different video frames. The basic information of the object specifically includes appearance information, motion state information and spatial distribution information. Determine whether the object information in the object recognition result meets the preset object association conditions; If so, the target objects are associated and organized to construct a corresponding object relationship description structure. Based on the object relationship description structure, key objects that have an impact on the target monitoring area are selected, and corresponding dynamic object files are established for the key objects.

[0012] This invention also provides an abnormal event recognition system for intelligent security cameras, comprising: The calculation module is used to identify the object type within the target monitoring area based on the video data pre-collected by the camera, generate corresponding target behavior state data according to the target behavior feature parameters of the object type in a continuous time period, match and analyze the target behavior state data with a preset set of behavior benchmark features, and calculate the behavior deviation feature parameters in the current monitoring scene. The object type specifically includes people, vehicles and objects, and the target behavior feature parameters specifically include position change, movement speed, dwell time, activity frequency and interaction relationship. The judgment module is used to determine whether the behavior deviation feature parameter exceeds a preset abnormal deviation threshold; The execution module is used to, if the time exceeds the specified time period, split the continuous video segments of the video data into behavioral states according to the time order within the continuous time period, obtain the abnormal stage state of the target object, construct the target behavior evolution sequence of the target object based on the abnormal stage state, collect the transition relationship between the abnormal stage states, and obtain the corresponding abnormal behavior evolution path. The abnormal stage state specifically includes the state before the abnormal behavior occurs, the abnormal change process state, and the current abnormal state. The second judgment module is used to determine whether the evolution process of the abnormal behavior evolution path can match the preset abnormal event characteristics. The second execution module is used to mark abnormal candidate objects from the abnormal behavior evolution path if there is no match, extract the regular characteristics of the abnormal candidate objects, generate corresponding abnormal event description features, perform cluster analysis between the abnormal event description features and historical abnormal event data, identify undefined abnormal behaviors with similar evolution trends, classify the undefined abnormal behaviors into preset abnormal event clusters, generate corresponding abnormal labels for the abnormal event clusters, and dynamically expand the abnormal event identification range of the monitoring environment through the abnormal labels. The regular characteristics specifically include behavioral change patterns, spatial activity patterns, and target interaction patterns.

[0013] Furthermore, it also includes: The generation module is used to perform continuous temporal correlation on the risk video clips based on anomaly deviation analysis, fuse the change information of the target object at different time nodes, and generate the continuous behavioral change trajectory of the target object. The change information specifically includes changes in position, changes in movement direction, changes in behavior duration, and changes in interaction with surrounding objects. The third judgment module is used to determine whether the continuous behavior change trajectory meets the preset continuous evolution conditions; The third execution module is used to identify the changing nodes in different abnormal stages based on the correlation between the abnormal stage states, select the time node with the fastest increase in abnormality from the changing nodes, dynamically divide the time node as the inflection point of the abnormal event evolution, and extract the behavioral combination features not covered by the existing abnormal event categories based on the inflection point of the abnormal event evolution.

[0014] Furthermore, the execution module also includes: The generation unit is used to obtain the target object status information corresponding to each abnormal stage based on the abnormal stage status, and to perform association matching on the target object status information according to the temporal relationship between different abnormal stages to generate corresponding stage association information. The target object status information specifically includes the spatial region where the target object is located, the duration of the behavior, the changes in the activity range, and the changes in the environment. The judgment unit is used to determine whether adjacent abnormal stages in the stage association information meet the preset continuous change conditions; The execution unit is configured to, if so, collect state evolution features between adjacent abnormal stages, extract stage transition features of the target object during the behavior change process, identify the behavior evolution features of the target object between different abnormal stages based on the stage transition features, divide the key change nodes when the abnormal state evolves from the initial stage to the subsequent stage, establish a state association structure between abnormal stages through the key change nodes, and combine abnormal stages with continuous evolution relationships to generate a corresponding continuous abnormal state chain. The state evolution features specifically include the state change direction, change magnitude, and change trend, and the behavior evolution features specifically include behavior enhancement, behavior migration, and behavior reversal.

[0015] This invention provides a method and system for identifying abnormal events in intelligent security cameras, which has the following beneficial effects: This invention can dynamically analyze the abnormal evolution process of target objects based on continuous behavioral changes, spatial activity patterns, and inter-object interactions within a monitored area, without relying on pre-defined fixed abnormal event categories. It can discover unknown abnormal behaviors that traditional abnormal classification models cannot cover. By constructing abnormal behavior evolution paths, extracting the regular characteristics corresponding to undefined abnormal events, and performing cluster analysis in conjunction with historical abnormal event data, it can automatically summarize similar abnormal trends and generate abnormal labels, enabling the system to continuously expand the scope of abnormal identification. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the abnormal event recognition method for intelligent security cameras according to the present invention. Figure 2This is a structural block diagram of an embodiment of the abnormal event recognition system for intelligent security cameras of the present invention. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The realization of the purpose, functional features and advantages of the invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings.

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Appendix Figure 1 An abnormal event recognition method for an intelligent security camera according to one embodiment of the present invention includes: S1: Based on the video data pre-collected by the camera in the target monitoring area, identify the object type in the target monitoring area, generate corresponding target behavior state data according to the target behavior characteristic parameters of the object type in a continuous time period, match and analyze the target behavior state data with a preset set of behavior benchmark features, and calculate the behavior deviation characteristic parameters in the current monitoring scene. The object type specifically includes people, vehicles and objects, and the target behavior characteristic parameters specifically include position change, movement speed, dwell time, activity frequency and interaction relationship. S2: Determine whether the behavioral deviation feature parameter exceeds a preset abnormal deviation threshold; S3: If the time exceeds the limit, the video data is split into behavioral states in the continuous video segments according to the time order within the continuous time period to obtain the abnormal stage state of the target object. Based on the abnormal stage state, the target behavior evolution sequence of the target object is constructed, the transition relationship between the abnormal stage states is collected, and the corresponding abnormal behavior evolution path is obtained. The abnormal stage state specifically includes the state before the abnormal behavior occurs, the abnormal change process state, and the current abnormal state. S4: Determine whether the evolution process of the abnormal behavior evolution path can match the preset abnormal event characteristics; S5: If there is no match, then mark the abnormal candidate objects from the abnormal behavior evolution path, extract the regular characteristics of the abnormal candidate objects, generate corresponding abnormal event description features, and perform cluster analysis between the abnormal event description features and historical abnormal event data to identify undefined abnormal behaviors with similar evolution trends. Divide the undefined abnormal behaviors into preset abnormal event clusters, generate corresponding abnormal labels for the abnormal event clusters, and dynamically expand the abnormal event identification range of the monitoring environment through the abnormal labels. The regular characteristics specifically include behavioral change patterns, spatial activity patterns, and target interaction patterns.

[0020] In this embodiment, the system identifies the object types within the target monitoring area based on video data pre-collected by the camera. Specific object types include people, vehicles, and objects. Based on the target behavior characteristic parameters of different object types over continuous time periods, including position changes, movement speed, dwell time, activity frequency, and interaction relationships, corresponding target behavior state data is generated. This target behavior state data is then matched and analyzed against a pre-set set of behavioral benchmark features (which describes the normal operating state of the monitoring area). Behavioral deviation characteristic parameters in the current monitoring scene are calculated, and the system determines whether these deviation parameters exceed a pre-set abnormal deviation threshold to execute corresponding steps. For example, if the system determines that the behavioral deviation characteristic parameters in the current monitoring scene do not exceed the pre-set abnormal deviation threshold, the system considers the behavior state of each object within the current monitoring area to be within a preset normal activity range. The system updates the normal activity patterns corresponding to different time periods, different spatial areas, and different object types based on the current behavioral state data that does not exceed the abnormal deviation threshold, and readjusts the association weights of various target behavior characteristic parameters in the behavioral benchmark feature set. For example, if the system determines that the behavioral deviation characteristic parameters in the current monitoring scene exceed the pre-set abnormal deviation threshold, the system will proceed accordingly. When a pre-defined abnormal deviation threshold is set, the system considers the behavior of objects within the current monitoring area to be abnormal. The system then breaks down the behavior into behavioral states within consecutive video segments based on the temporal sequence of the video data, obtaining the abnormal stage states of the target objects. These abnormal stage states specifically include the state before the abnormal behavior occurs, the state during the abnormal change process, and the current abnormal state. Based on these different abnormal stage states, the system constructs the target object's target behavior evolution sequence, collects the transition relationships between different abnormal stage states, and obtains the corresponding abnormal behavior evolution path. By acquiring the state before the abnormal behavior occurs, the state during the abnormal change process, and the current abnormal state, the system can completely preserve the target object's behavior. The entire process of a target object transitioning from a normal state to an abnormal state allows for a more accurate understanding of the causes and development of abnormal behavior, improving the reliability of abnormal event identification in complex monitoring environments. Furthermore, when new, undefined abnormal behaviors appear within the monitoring area, the system can still extract event features based on the state transition relationships and behavioral evolution patterns between abnormal stages. This provides a data foundation for subsequent identification of unknown abnormal events, generation of abnormal tags, and expansion of the abnormal identification scope, enhancing the adaptability of intelligent security systems to different application scenarios and complex environmental changes. The system then determines whether the evolutionary path of the abnormal behavior matches preset abnormal event features, and executes the corresponding steps accordingly.For example, when the system determines that the evolution of an abnormal behavior path matches preset abnormal event characteristics, the system considers the current abnormal behavior not to be an unknown anomaly, but rather a known risk event within the preset abnormal event range. Based on the matched abnormal event characteristics, the system extracts the key behavioral nodes corresponding to the abnormal event, including the abnormal behavior trigger time, the anomaly occurrence area, the associated target object, and the anomaly persistence state. Combining this with real-time changes in the current abnormal behavior evolution path, the system locates the development stage of the abnormal event and sends the abnormal event description information to the monitoring and management terminal. Conversely, when the system determines that the evolution of an abnormal behavior path does not match preset abnormal event characteristics, the system considers the current abnormal behavior to be an unknown anomaly. The system marks abnormal candidate objects from the abnormal behavior evolution path and extracts the regular characteristics of these candidate objects. These regular characteristics specifically include behavioral change patterns and spatial activity patterns. The system identifies patterns and interaction patterns between targets, generates corresponding descriptive features for abnormal events, and performs cluster analysis on these features with historical abnormal event data to identify undefined abnormal behaviors with similar evolutionary trends. These undefined abnormal behaviors are then assigned to pre-defined abnormal event clusters, and corresponding abnormal labels are generated for each cluster. These labels dynamically expand the scope of abnormal event identification in the monitoring environment. The system is no longer limited to identifying pre-defined abnormal events such as falls, intrusions, and gatherings; it can proactively discover potential risks based on new behavioral changes that occur during actual monitoring. It possesses the ability to perceive unknown abnormal events, improving the scope of anomaly detection in complex application scenarios. Furthermore, by integrating information on changes in target objects in time, space, and interaction relationships, it generates more complete descriptive features for abnormal events, enabling the system to accurately characterize the essential features of undefined abnormal behaviors and avoiding the problem of being unable to identify them due to a lack of pre-defined categories.

[0021] It should be noted that, based on the temporal order within a continuous time period, the continuous video segments of the video data are divided into behavioral states to obtain the abnormal stage states of the target object. Based on these abnormal stage states, a target behavior evolution sequence of the target object is constructed, and the transition relationships between these abnormal stage states are collected to obtain the corresponding abnormal behavior evolution path. Specifically: The system first performs correlation analysis on the state changes of the target object in consecutive video frames according to the time sequence before and after the anomaly occurs. It identifies key change nodes in the process of the target object transitioning from a normal state to an abnormal state, and divides the consecutive video segments into multiple behavioral stages based on these change nodes. Among them, the state before the abnormal behavior occurs describes the basic activity state of the target object before the anomaly occurs, such as the target object's location area, movement mode, dwell time, and normal interaction with surrounding objects; the abnormal change process state describes the process of the target object gradually undergoing abnormal changes from a normal state, such as sudden changes in movement speed, expansion of activity range, increase in dwell time, or abnormal proximity to other objects; the current abnormal state describes the abnormal behavior results currently exhibited by the target object, such as entering a restricted area, continuous wandering, abnormal gathering, or abnormal interaction with other objects. Furthermore, based on the temporal sequence between different abnormal phase states, the system correlates the target behavior feature parameters in each abnormal phase, extracts the state change relationship of the target object between different phases, and arranges the states of each phase according to the order of occurrence of the abnormal phases to generate the target behavior evolution sequence corresponding to the target object. This target behavior evolution sequence is not simply a record of the time of occurrence of the abnormality, but is used to describe the continuous change process of the target object's behavior state from the initial change to the formation of the abnormality. For example, in the scenario of a person illegally entering an area, the system can first identify the state of the person walking normally in the public area, then identify the state of the person gradually approaching the restricted area and stopping to observe, and finally identify the abnormal state after entering the restricted area. Through the arrangement of the above continuous states, a behavior evolution sequence of "normal approach state - abnormal preparation state - illegal entry state" is formed. After obtaining the target behavior evolution sequence, the system further collects the transition relationships between adjacent abnormal stage states, analyzes the changes in the target object's position, movement speed, dwell state, and interaction objects between different stages, determines the change pattern of abnormal behavior when it transitions from one stage to another, and generates corresponding abnormal behavior evolution paths based on the degree of correlation between each stage. For example, in a person's fall event, the system can identify the process of a person changing from a normal standing state to a state of imbalance, and then to a state of lying down, and generate an abnormal behavior evolution path of "normal activity state - abnormal imbalance state - fall state" based on the transition relationships between different states. In this way, the system can extend from the analysis of a single abnormal result to the analysis of the abnormal event formation process, providing complete data basis for subsequent identification of unknown abnormal events, expansion of abnormal types, and judgment of risk trends.

[0022] It should be added that, from the evolution path of the abnormal behavior, candidate abnormal objects are marked, the regular characteristics of the candidate abnormal objects are extracted, corresponding abnormal event description features are generated, and cluster analysis is performed between the abnormal event description features and historical abnormal event data to identify undefined abnormal behaviors with similar evolutionary trends. These undefined abnormal behaviors are then assigned to preset abnormal event clusters, and corresponding abnormal labels are generated for each cluster. Through these abnormal labels, the scope of abnormal event identification in the monitoring environment is dynamically expanded. Specifically: Based on the changes in the state at each stage of the abnormal behavior evolution path, the system analyzes the degree of participation of different target objects in the occurrence, development, and continuation of the abnormality, and identifies objects that continuously affect the changes in the abnormal state as candidate objects for anomalies. For example, in a shopping mall monitoring scenario, if the system finds that a person continuously approaches a closed area at multiple time stages, and the changes in the person's position, dwell time, and interaction with the surrounding people all contribute to the formation of an abnormal state, then the person is marked as a candidate object for anomalies, rather than the people passing by normally as the objects of anomaly analysis. Furthermore, the system analyzes the performance of anomalous candidate objects during the evolution of anomalous behavior, extracting corresponding regular characteristics. These regular characteristics include behavioral change patterns, spatial activity patterns, and target interaction patterns. Behavioral change patterns describe the changes in the behavioral state of an anomalous candidate object over a continuous period, such as a gradual increase in movement speed, a continuous increase in dwell time, or a change in behavioral direction. Spatial activity patterns describe the changes in the location distribution and activity range of the anomalous candidate object within the monitored area, such as gradually moving from an open area to a restricted area, or repeatedly moving around a certain area. Target interaction patterns describe the changes in the association between the anomalous candidate object and other people, vehicles, or objects, such as abnormally approaching a target, frequent interaction, or following other objects. By fusing these regular characteristics, the system can form anomalous event description feature to describe the formation process of unknown anomalous behavior. After generating the descriptive features of the abnormal events, the system further performs cluster analysis on these features and historical abnormal event data. By comparing the behavioral change patterns, spatial activity patterns, and target interaction patterns among different abnormal events, the system determines whether there are similar evolutionary trends between the current unknown abnormal behavior and historical abnormal events. If multiple unknown abnormal behaviors have high similarity in terms of abnormal formation process, stage change relationship, and target activity patterns, the system classifies these unknown abnormal behaviors into the corresponding abnormal event clusters. For example, if the system first discovers that a person repeatedly enters the area around a warehouse at night and continues to observe, but this behavior does not belong to existing abnormal categories such as theft or trespassing, and subsequent monitoring reveals that different people all exhibit a similar evolutionary process of "approaching the warehouse - staying for a long time - observing the environment - leaving," then the system can classify this type of behavior into a new abnormal event cluster. Furthermore, based on the common evolutionary patterns of unknown abnormal behaviors within an abnormal event cluster, the system generates corresponding abnormal tags for that cluster and adds these tags to the abnormal event identification range of the monitoring environment. This allows the system to quickly identify similar behaviors when they reappear during subsequent monitoring based on the newly added abnormal tags. Simultaneously, as monitoring data continues to accumulate, the system can continuously update the behavioral patterns and descriptive features within the abnormal event cluster, enabling the anomaly identification capability to expand with changes in the actual application environment. This reduces the reliance of intelligent security cameras on preset anomaly categories and improves the proactive detection capability of unknown abnormal events.

[0023] In this embodiment, after step S3 of splitting the continuous video segments of the video data into behavioral states according to the time sequence within a continuous time period to obtain the abnormal stage state of the target object, the method further includes: S301: Based on the risk video segment of the anomaly deviation analysis, the target object in the risk video segment is continuously correlated over time, and the change information of the target object at different time nodes is fused to generate the continuous behavior change trajectory of the target object. The change information specifically includes changes in position, changes in movement direction, changes in behavior duration, and changes in interaction with surrounding objects. S302: Determine whether the continuous behavior change trajectory meets the preset continuous evolution conditions; S303: If so, then based on the correlation between the states of the abnormal stages, identify the changing nodes in different abnormal stages, select the time node with the fastest increase in abnormality from the changing nodes, dynamically divide the time node as the inflection point of the abnormal event evolution, and extract the behavioral combination features that are not covered by the existing abnormal event categories based on the inflection point of the abnormal event evolution.

[0024] In this embodiment, the system performs continuous temporal correlation on risk video segments based on anomaly deviation analysis ("anomaly deviation analysis" originates from the aforementioned steps of matching real-time acquired target behavior feature parameters with a normal behavior baseline feature set to calculate behavior deviation parameters, and determining risk video segments based on behavior deviation parameters exceeding a preset threshold). It fuses the change information of the target object at different time points, specifically including changes in position, direction of movement, duration of behavior, and interaction with surrounding objects, generating a continuous behavior change trajectory for the target object. The system then determines whether this continuous behavior change trajectory meets preset continuous evolution conditions to execute corresponding steps. For example, if the system determines that the continuous behavior change trajectory of the target object cannot meet the preset continuous evolution conditions, the system considers that although the current target object exhibits a certain degree of behavior deviation, the behavior change process has not formed a continuous state transition relationship. The system analyzes the influencing factors causing the interruption of behavior evolution based on the discontinuous change nodes in the continuous behavior change trajectory, classifies and marks the current behavior change state, generates corresponding abnormal fluctuation record information, and reduces the behavior. The system identifies the corresponding abnormal risk level for each state to avoid misjudging occasional behavior as a continuous abnormal event. For example, when the system determines that the continuous behavioral change trajectory of a target object meets the pre-set continuous evolution conditions, the system considers the abnormal change of the target object to be a frequent state. The system identifies the change nodes in the abnormal stage based on the correlation between different abnormal stage states, selects the time node with the fastest increase in abnormality from these change nodes, and dynamically divides each time node into an inflection point of abnormal event evolution. Based on these inflection points of abnormal event evolution, the system extracts behavioral combination features not covered by existing abnormal event categories. By dynamically dividing the inflection points of abnormal event evolution, the system can grasp the development rhythm and risk change trend of abnormal behavior. It can not only understand the current abnormal state, but also analyze the key impact stages in the formation process of abnormal events, providing more accurate data for subsequent risk prediction and early intervention. At the same time, it can discover new behavioral combination patterns from the continuous evolution process, such as new correlations formed between changes in the target object's activity range, changes in interaction relationships, and changes in behavioral intensity, and use them as the basis for identifying unknown abnormal events. This improves the ability of intelligent security cameras to actively detect new and complex abnormal behaviors and further expands the scope of abnormal event identification.

[0025] It should be noted that, based on the correlation between the states of the aforementioned abnormal stages, change nodes in different abnormal stages are identified. From these change nodes, the time nodes with the fastest increase in abnormality are selected, and these time nodes are dynamically classified as inflection points in the evolution of abnormal events. Based on these inflection points, behavioral combination features not covered by existing abnormal event categories are extracted, specifically: The system performs correlation analysis on the target's behavioral characteristic parameters in each abnormal stage, including changes in the target object's position, movement state, dwell time, activity range, and interaction with surrounding objects. It determines whether there is a significant state transition between adjacent abnormal stages and marks the time nodes with more obvious state changes as change nodes. For example, in an abnormal wandering event, the system can identify the process of the target object entering a state of repeated back-and-forth movement from a normal walking state, and then entering a state of prolonged dwelling. The locations where the direction of movement changes, the dwell time increases, and the activity area shrinks can all be used as change nodes between abnormal stages. Furthermore, the system analyzes the degree of anomaly of the identified multiple change nodes. Based on the magnitude, duration, and scope of influence of the behavioral changes at different time nodes, it calculates the abnormal growth of each change node and selects the time node with the fastest growth in anomaly degree from multiple change nodes, identifying this time node as the inflection point of the abnormal event evolution. This inflection point of the abnormal event evolution is used to represent the key stage in the transformation of abnormal behavior from a normal deviation state to a clearly risky state. For example, in the scenario of abnormal vehicle parking, the initial entry of the vehicle into the parking area is normal behavior. When the vehicle gradually changes from a short stop to a long stay, and there are frequent approaches and observations by people around, the system will identify the time point of rapid change in the vehicle's parking state and use it as the inflection point of the abnormal event evolution for in-depth analysis of subsequent behavioral development. After identifying the inflection point in the evolution of an anomaly, the system extracts behavioral combination features not covered by existing anomaly categories based on the behavioral changes before and after the inflection point. Specifically, the system analyzes the joint changes of the target object in multiple dimensions before and after the inflection point, such as the combination of location migration patterns, changes in movement state, dwelling patterns, and changes in object interaction, rather than analyzing a single abnormal behavioral feature. In this way, new behavioral patterns not defined in traditional anomaly category libraries can be discovered. For example, if the system finds that a person exhibits a continuous behavioral combination of "approaching a sensitive area multiple times—staying briefly to observe—approaching alternately with other people—leaving quickly," although this behavior does not conform to existing single anomaly categories such as intrusion or loitering, its overall evolution process has obvious risk characteristics. The system can then extract this behavioral combination feature as a basis for subsequent identification of unknown anomalies and expansion of anomaly categories. Through the above process, intelligent security cameras can shift from identifying single abnormal behaviors to analyzing the development patterns of anomalies, improving their ability to proactively detect complex and unknown anomalies.

[0026] In this embodiment, step S3, which involves constructing the target behavior evolution sequence of the target object based on the abnormal stage states, collecting the transition relationships between the abnormal stage states, and obtaining the corresponding abnormal behavior evolution path, further includes: S31: Based on the abnormal stage state, obtain the target object state information corresponding to each abnormal stage, and perform association matching on the target object state information according to the time sequence relationship between different abnormal stages to generate corresponding stage association information. The target object state information specifically includes the spatial area where the target object is located, the duration of the behavior, the change of the activity range, and the change of the environment. S32: Determine whether adjacent abnormal stages in the stage association information meet the preset continuous change condition; S33: If so, then collect the state evolution features between adjacent abnormal stages, extract the stage transition features of the target object in the process of behavioral change, identify the behavioral evolution features of the target object between different abnormal stages based on the stage transition features, divide the key change nodes when the abnormal state evolves from the initial stage to the subsequent stage, establish the state association structure between abnormal stages through the key change nodes, combine abnormal stages with continuous evolution relationship to generate the corresponding continuous abnormal state chain, wherein the state evolution features specifically include the state change direction, change amplitude and change trend, and the behavioral evolution features specifically include behavior enhancement, behavior migration and behavior turning situation.

[0027] In this embodiment, the system obtains the target object state information corresponding to each abnormal stage based on these abnormal stage states. The target object state information specifically includes the target object's spatial region, behavior duration, changes in activity range, and environmental changes. Based on the temporal relationship between different abnormal stages, the system performs correlation matching on this target object state information to generate corresponding stage correlation information. Then, the system determines whether adjacent abnormal stages in the stage correlation information meet a pre-set continuous change condition (the continuous change condition is used to determine whether there is a state continuity relationship between different abnormal stages) to execute the corresponding steps. For example, when the system determines that there is no adjacent abnormal stage in the stage correlation information... When the conditions for continuous change are not met, the system assumes that there is no obvious state continuity between adjacent abnormal stages and that the target object has not formed a stable development trend between different abnormal stages. Based on the stage association information that fails to meet the conditions for continuous change, the system identifies the location of the break between adjacent abnormal stages, analyzes the cause of the interruption in state change, marks the abnormal stages without continuous evolution as independent abnormal observation segments, and records the corresponding time location, spatial region, and target object state information as the data basis for subsequent abnormal behavior analysis. For example, when the system determines that adjacent abnormal stages in the stage association information can meet the pre-set conditions for continuous change, the system... The system assumes a high degree of correlation between current abnormal stages and a state continuity relationship between adjacent abnormal stages. It collects state evolution characteristics between adjacent abnormal stages, specifically including the direction, magnitude, and trend of state changes. It extracts stage transition characteristics for each target object during behavioral changes. Based on these stage transition characteristics, it identifies the behavioral evolution characteristics of each target object between different abnormal stages, specifically including behavioral enhancement, behavioral migration, and behavioral shift. It identifies key change nodes in the evolution of abnormal states from the initial stage to subsequent stages. Through these key change nodes, it establishes a state correlation structure between abnormal stages, identifying those with continuous evolutionary relationships. Abnormal stages are combined to generate corresponding continuous abnormal state chains. By analyzing the direction, magnitude, and trend of state changes, the system can accurately grasp the development pattern of a target object from its initial abnormal state to subsequent abnormal states. This avoids the problem of traditional security monitoring relying solely on a single abnormal moment, which leads to the inability to identify the abnormal development process. It improves the completeness and accuracy of abnormal event identification. At the same time, it identifies key stages in which the target object's behavior develops from a slight deviation to a high-risk state, such as a gradual expansion of its activity range, a gradual deviation of its movement direction from the normal area, or an increasing degree of interaction with surrounding objects. This enables dynamic analysis of the development trend of abnormal events, providing a basis for subsequent risk prediction and early intervention.

[0028] It should be noted that the process involves collecting state evolution features between adjacent abnormal stages, extracting stage transition features of the target object during behavioral changes, identifying behavioral evolution features of the target object between different abnormal stages based on these stage transition features, dividing the key change nodes when the abnormal state evolves from the initial stage to subsequent stages, establishing a state association structure between abnormal stages through these key change nodes, and combining abnormal stages with continuous evolutionary relationships to generate corresponding continuous abnormal state chains. Specifically: The system acquires information on the direction of change in the target object's position, the degree of change in its activity range, changes in its movement state, changes in its dwell time, and changes in its interaction with surrounding objects between adjacent abnormal stages. Based on the differences between two consecutive abnormal stages, the system determines the pattern of change in the target object's state as it transitions from one stage to another. For example, if a person is walking normally before the anomaly occurs, and as they gradually approach the restricted area, the system analyzes whether the person's movement direction has shifted, whether their movement speed has decreased, and whether their dwell time has increased, using these changes as characteristics of the state evolution between adjacent abnormal stages. Furthermore, based on the state evolution characteristics, the system extracts the stage transition characteristics of the target object during the behavioral change process. These stage transition characteristics describe the specific changes the target object makes when it transitions from one abnormal stage to another, including the process of the abnormal state changing from weak to strong, the process of migrating the activity area, and the process of changing the behavioral purpose. For example, in a scenario of abnormal loitering, the target object may initially briefly stay in a certain area, then gradually expand its activity range and return to the same area multiple times, and finally remain there continuously observing the surrounding environment. By analyzing the transition relationships between these stages, the system can determine the behavioral change process of the target object transitioning from a "brief stay state" to a "continuous attention state." After obtaining the phase transition characteristics, the system further identifies the behavioral evolution characteristics of the target object between different abnormal stages, and analyzes the enhancement, migration, and turning of abnormal behavior during the development process. Among them, behavioral enhancement describes the situation where the degree of abnormality gradually increases, such as increased dwell time, increased movement frequency, or expanded activity range; behavioral migration describes the situation where the target object's activity location changes, such as gradually moving from a public area to a restricted area; behavioral turning describes the situation where the target object's behavioral purpose or activity direction changes, such as changing from normal passage to repeated activity around a certain area. By identifying the above behavioral evolution characteristics, the system can determine the development direction of abnormal behavior, rather than only focusing on the abnormal behavior at the current moment. Subsequently, based on the changes in behavioral evolution characteristics, the system identifies key change nodes in the evolution of abnormal states from the initial stage to subsequent stages. Specifically, the system analyzes the locations where the degree of abnormality changes significantly during the transition between each stage, identifying the time points that lead to a rapid increase in the abnormal state, a significant change in behavioral patterns, or an increase in risk level as key change nodes. For example, in an abnormal vehicle parking event, a vehicle may be exhibiting normal parking behavior when it first enters the area, but when the vehicle changes from a short-term parking to a long-term parking, and there are changes such as people approaching to observe or frequent movement, this transition time point can be considered a key change node in the evolution of the abnormal event. Finally, the system utilizes the aforementioned key change nodes to establish a state association structure between abnormal stages, combining abnormal stages with continuous evolutionary relationships to generate corresponding continuous abnormal state chains. These continuous abnormal state chains can describe the complete process of a target object from the generation and development of an anomaly to its formation in chronological order. For example, for an illegal entry into a restricted area event, the system can combine the "approaching restricted area state - observation and stay state - entering restricted area state - continuous activity state" into a continuous abnormal state chain according to the stage association relationship.

[0029] In this embodiment, step S5, which involves marking abnormal candidate objects from the abnormal behavior evolution path, extracting the regular characteristics of the abnormal candidate objects, and generating corresponding abnormal event description features, further includes: S51: Based on the abnormal behavior evolution path, obtain the change information set corresponding to each abnormal stage, perform correlation analysis on different state nodes in the abnormal behavior evolution path, construct the change relationship between each state node, and according to the change relationship, filter target objects with continuous change characteristics and generate a corresponding candidate object set. The change information set specifically includes target object state change information, spatial activity change information, and associated object change information. S52: Determine whether the target object in the candidate object set satisfies the preset abnormal evolution association condition; S53: If so, obtain the behavioral performance information of the target object in different evolutionary stages, collect the change patterns of the target object in different dimensions based on the behavioral performance information, construct the behavioral description structure of the target object based on the correlation between the change patterns, and divide the change combination features of existing abnormal categories through the behavioral description structure. The different dimensions specifically include time dimension, space dimension and interaction dimension.

[0030] In this embodiment, based on the abnormal behavior evolution path, the system obtains a set of change information corresponding to each abnormal stage. Specifically, the set of change information includes target object state change information, spatial activity change information, and related object change information. The system performs correlation analysis on different state nodes in the abnormal behavior evolution path, constructs the change relationships between each state node, and filters target objects with continuous change characteristics based on different change relationships, generating a corresponding candidate object set. Then, the system determines whether the target objects in the candidate object set meet the pre-set abnormal evolution correlation conditions (the abnormal evolution correlation conditions are used to determine whether the target object is in the process of abnormal behavior development). The system continuously participates in abnormal state changes, including the number of abnormal stages a target object participates in, the continuity of state changes, and changes in association with other objects, in order to execute corresponding steps. For example, when the system determines that a target object in the candidate object set cannot meet the pre-set abnormal evolution association conditions, the system will consider that although the current candidate object appears in the abnormal behavior evolution path, it has not continuously participated in the development process of the abnormal state. The system will analyze the specific reasons why the target object does not meet the abnormal evolution association conditions based on its participation in the abnormal behavior evolution path, and re-screen the target objects in the candidate object set to ensure that those that cannot continuously influence the abnormal state change are removed. Objects exhibiting constant state changes are marked as low-association objects and removed from subsequent anomalous event evolution analysis to avoid bias in anomalous event identification results due to the involvement of irrelevant objects. For example, when the system determines that a target object in the candidate object set meets pre-defined anomalous evolution association conditions, the system considers the target object to have a high degree of association with the anomalous event and requires it to be the subject of anomalous event analysis. The system acquires the target object's behavioral performance information at different evolution stages and, based on this information, collects the target object's change patterns across different dimensions, including time, space, and interaction. Based on the correlation between these change patterns, the system constructs a behavioral description structure for the target object. Through this behavioral description structure, it classifies the change combination features of existing anomaly categories. The system can focus on key objects that continuously participate in abnormal state changes and analyze their complete behavioral processes, thereby improving the accuracy of anomaly location in multi-target monitoring environments and reducing false alarms caused by target mixing. At the same time, when a target object exhibits a new combination of behaviors not covered by existing categories, the system can still identify its correlation with historical anomaly patterns based on the behavioral description structure, thereby enhancing the adaptability of intelligent security cameras to complex environments and unknown anomaly events.

[0031] It should be noted that the process involves acquiring behavioral information of the target object at different evolutionary stages, collecting the change patterns of the target object across different dimensions based on this behavioral information, constructing a behavioral description structure of the target object based on the correlation between these change patterns, and then using this behavioral description structure to classify the change combination features of existing anomaly categories. Specifically: The system extracts the target object's state information at each stage of the abnormal behavior evolution path, including the target object's location area, movement trajectory, movement speed, dwell time, activity range, and interaction with surrounding objects. It then organizes this behavioral information in chronological order to form a complete record of the target object's stage behavior. For example, in a scenario where a person abnormally approaches an important area, the system not only acquires the person's behavior of finally entering the restricted area, but also records the entire process of the person moving from a distance to the target area, gradually reducing their movement speed, stopping to observe, and interacting with surrounding people. Furthermore, based on the acquired behavioral information, the system collects the changing patterns of the target object during the abnormal evolution process from different dimensions; specifically, these include changes in the time, space, and interaction dimensions. The time dimension is used to analyze the chronological order and continuous changes in the target object's behavior, such as whether the dwell time gradually increases or whether the abnormal behavior continues. The space dimension is used to analyze changes in the target object's location and movement range, such as whether it gradually moves from a normal activity area to a sensitive area, or whether it repeatedly moves or stays in a fixed area. The interaction dimension is used to analyze changes in the relationship between the target object and other people, vehicles, or objects, such as whether abnormal approaching, following, or collaborative behaviors occur. By analyzing multiple dimensions, the formation process of the target object's abnormal behavior can be described from different perspectives. Subsequently, based on the correlation between the change patterns of different dimensions, the system integrates the information on time, space, and interaction changes to construct a behavioral description structure corresponding to the target object. This behavioral description structure is used to represent the comprehensive behavioral pattern of the target object in the development process of abnormal events, transforming the originally independent behavioral change information into a logically related overall description. For example, if the system finds that a person in the nighttime monitoring area exhibits a continuous change process of "gradually approaching the warehouse area - stopping to observe multiple times - briefly interacting with another person - entering a non-open area", then it correlates these time changes, spatial changes, and interaction changes to form the behavioral description structure corresponding to that person. Finally, the system uses the behavior description structure to classify and extract the combination features of changes in existing anomaly categories, and summarizes the representative combinations of behavior changes in different anomaly events. For example, for illegal entry events, the system can extract the combination features of "approaching the restricted area - breaking through the area boundary - staying continuously" from existing anomaly categories; for abnormal loitering events, it can extract the combination features of "moving within a fixed range - repeatedly returning to the target location - staying for a long time".

[0032] In this embodiment, step S2, which determines whether the behavioral deviation feature parameter exceeds a preset abnormal deviation threshold, further includes: S21: Based on the aforementioned behavioral deviation feature parameters, establish a corresponding deviation correlation structure; S22: Determine whether the deviation factors in the deviation association structure meet the preset association evolution conditions; S23: If so, then based on the deviation association structure, obtain the main influencing factors in the target object's behavior change process, collect the evolution order among the main influencing factors, and divide the abnormal changes generated by the target object in a continuous time period into stages according to the evolution order, and identify the deviation process caused by behavior change. The deviation process specifically includes the local deviation process caused by a single behavior change and the composite deviation process formed by the combined action of multiple behavior factors.

[0033] In this embodiment, the system establishes a corresponding deviation correlation structure based on these behavioral deviation feature parameters (the deviation correlation structure is used to describe the mutual influence relationship between various behavioral factors during the abnormal change of the target object). Then, the system determines whether the deviation factors in the deviation correlation structure meet the preset correlation evolution conditions (the correlation evolution conditions are used to determine whether there is a synchronous change, continuous change, or mutual influence relationship between different deviation factors), and executes the corresponding steps accordingly. For example, when the system determines that the deviation factors in the deviation correlation structure cannot meet the preset correlation evolution conditions, the system will consider that although each deviation factor may exhibit a certain degree of abnormal fluctuation, it has not formed a mutual influence and co-evolution process. The system will split the deviation factors that cannot form a correlation evolution relationship according to the degree of correlation between each deviation factor in the deviation correlation structure, identify the independent change source corresponding to each deviation factor, analyze each independent deviation factor separately, extract its occurrence time, area of ​​action, change amplitude, and duration, and generate corresponding deviation status record information. For example, when the system determines that the deviation factors in the deviation correlation structure can meet the preset correlation evolution conditions, the system will consider that in the current monitoring scenario... The system identifies the main influencing factors influencing the behavior deviations of the target object, based on the correlation structure. It then collects the evolutionary sequence of these factors and divides the abnormal changes occurring within the target object over a continuous period into stages. This process identifies deviations caused by behavioral changes, including local deviations caused by single behavioral changes and complex deviations resulting from the combined effects of multiple behavioral factors. The system no longer focuses solely on the abnormal outcome but analyzes the temporal relationship and interaction of various influencing factors. For example, it examines how factors such as an expanded activity range, a change in movement direction, and increased interaction with surrounding objects collectively drive the formation of abnormal states. This allows for a more accurate assessment of the development trend of abnormal events, providing a more complete process basis for analyzing unknown abnormal behaviors. Furthermore, for anomalies caused by a single factor, the system can quickly pinpoint the specific source of influence. For complex anomalies formed by the combined effects of multiple factors, the system analyzes the correlation and collaborative changes between these factors. This enhances the ability of intelligent security cameras to detect unknown abnormal events in complex scenarios and reduces the limitations of traditional fixed-rule-based identification methods.

[0034] It should be noted that, based on the aforementioned deviation correlation structure, the main influencing factors in the target object's behavioral change process are obtained, the evolutionary sequence among these main influencing factors is collected, and based on this evolutionary sequence, the abnormal changes generated by the target object within a continuous time period are divided into stages to identify the deviation process caused by behavioral changes. Specifically: The system analyzes the degree of influence of different deviation factors in the formation of abnormal behavior. Based on the magnitude of change, duration, and correlation with other factors of each deviation factor, the system screens out the factors that have the main influence on the abnormal state of the target object. For example, in the scenario where a person abnormally approaches a restricted area, the system finds several behavioral change factors such as the target object's change of movement direction, gradual reduction of activity range, increase in dwell time, and reduction in interaction with surrounding people. Among these, the continuous approach to the restricted area and the continuous increase in dwell time have a greater impact on the formation of the abnormal state. Therefore, these two factors are selected as the main influencing factors for subsequent analysis. Furthermore, the system collects the evolutionary sequence of the extracted main influencing factors during the development of abnormal behavior, and analyzes the sequential relationship and mutual influence process of different factors. Specifically, based on video data over a continuous period of time, the system correlates the change time nodes corresponding to each main influencing factor to determine which factor first caused the abnormal behavior, and which factors gradually accumulated to form it. For example, in a scenario of abnormal vehicle parking, the system may first detect that the vehicle has entered an abnormal parking area, then find that the vehicle's parking time continues to increase, and finally detect that the people inside the vehicle are frequently observing the surrounding environment. In this case, the system can determine the evolutionary sequence of factors: "entering an abnormal area - continuous parking - abnormal observation". Subsequently, the system divides the abnormal change process of the target object over a continuous period of time into stages based on the evolutionary order among the main influencing factors. Specifically, according to the temporal relationship between the appearance and development of abnormal factors, the system divides the continuous abnormal change process into multiple stages, including the initial stage of abnormal change, the development stage of abnormal state, and the formation stage of abnormal state. In the initial stage of abnormal change, the target object may only exhibit a single behavioral deviation. In the development stage of abnormal state, multiple influencing factors begin to show correlated changes. In the formation stage of abnormal state, the target object has already exhibited more obvious abnormal behavioral characteristics. For example, for abnormal loitering behavior of personnel, the system can classify "first entry into the area of ​​interest" as the initial deviation stage, "multiple round trips and extended stay time" as the development stage, and "continuous stay and abnormal interaction with surrounding objects" as the formation stage. Finally, based on the changing relationships of the main influencing factors in different stages, the system identifies the deviation process caused by behavioral changes and determines whether the current abnormal change is a local deviation caused by a single behavioral factor or a complex deviation process formed by the combined effect of multiple behavioral factors. For local deviations caused by a single behavioral change, such as a person leaving after a short stay, the system can identify it as a low-association deviation state. For complex deviations formed by the combined effect of multiple factors, such as a person continuously approaching a sensitive area, increasing the stay time and accompanied by abnormal interactive behavior, the system can identify it as an abnormal deviation process with a development trend.

[0035] In this embodiment, step S4, which determines whether the evolution process of the abnormal behavior evolution path matches the preset abnormal event characteristics, further includes: S41: Based on the abnormal behavior evolution path, obtain the correlation strength between each evolution stage, and according to the correlation strength, analyze the evolution structure in the abnormal behavior evolution path to generate path structure parameters of the abnormal behavior development law; S42: Determine whether the path structure parameters meet the preset path stability conditions; S43: If so, then based on the stage association relationship in the abnormal behavior evolution path, extract the evolution nodes in the process of the abnormal behavior of the target object changing from the initial state to the subsequent state, identify the change connection relationship between the evolution nodes, determine the development links in the abnormal behavior development process, and construct the abnormal behavior evolution logic chain, wherein the development links specifically include the behavior triggering link, the state transition link, and the risk diffusion link.

[0036] In this embodiment, the system obtains the correlation strength between each evolution stage based on the abnormal behavior evolution path. According to different correlation strengths, it analyzes the evolution structure in the abnormal behavior evolution path to generate path structure parameters of the abnormal behavior development pattern. Then, the system determines whether these path structure parameters meet pre-set path stability conditions (path stability conditions are used to determine whether there is a continuous state change relationship and a clear development direction during the abnormal behavior evolution process), and executes corresponding steps accordingly. For example, when the system determines that the path structure parameters of the abnormal behavior development pattern cannot meet the pre-set path stability conditions, the system will consider that the abnormal state of the target object may have stage jumps, interruptions in change, or random fluctuations, and cannot form a stable abnormal development trend. The system will analyze the abnormal behavior evolution path based on the parts of the path structure parameters that do not meet the stability conditions, identify the influencing factors that cause path instability, re-divide the unstable stages in the abnormal behavior evolution path, separate the abnormal stages that cannot form a continuous evolution relationship, and retain the time information, spatial information, and behavior change information corresponding to each stage to generate independent abnormal state records. For example, when the system determines that the path structure parameters of the abnormal behavior development pattern can meet the pre-set path stability conditions, it will execute corresponding steps accordingly. If the pre-defined path stability conditions are met, the system will consider that there are connections between the various evolutionary stages in the current abnormal behavior evolution path, forming a stable abnormal development trend. Based on the stage connections in the abnormal behavior evolution path, the system will extract the evolutionary nodes in the process of the target object's abnormal behavior changing from the initial state to the subsequent state, identify the connection relationships between these evolutionary nodes, and determine the development links in the abnormal behavior development process. The development links specifically include the behavior triggering link, the state transition link, and the risk diffusion link, constructing an abnormal behavior evolution logic chain. The system can identify the triggering factors, state change process, and risk expansion path in the process of abnormal behavior from generation to escalation. For example, the target object initially shows abnormal activity tendencies, then the behavior state undergoes a significant transition, and finally the scope of influence gradually expands. This enables a deep understanding of the development law of abnormal events, providing a basis for early warning and risk intervention. At the same time, the system can record the current performance of abnormal behavior and analyze the causes, development trends, and potential impact range of abnormal events. This allows intelligent security cameras to break away from the dependence on fixed abnormal categories and actively identify unknown abnormal events based on the behavior evolution law, improving the system's adaptability to complex monitoring environments and new risk behaviors.

[0037] It should be noted that, based on the stage relationships in the abnormal behavior evolution path, the evolution nodes in the process of the target object's abnormal behavior changing from the initial state to subsequent states are extracted, the connection relationships between the evolution nodes are identified, the development stages in the abnormal behavior development process are determined, and the abnormal behavior evolution logic chain is constructed, specifically as follows: The system analyzes the changes in the target object's state during each abnormal stage, including changes in behavioral state, spatial location, activity range, and interaction with surrounding objects, identifying key time points that lead to changes in the abnormal state. These evolutionary nodes do not simply correspond to a single point in time, but rather represent critical stages where the target object's behavioral pattern undergoes a significant change. For example, in an event of abnormal entry into a restricted area, the target object may initially only approach the restricted area, then remain there for an extended period of observation, and finally enter the restricted area and continue its activity. The system can identify "approaching the restricted area," "remaining there for observation," and "entering the restricted area" as key nodes in the evolution of the abnormal behavior. Furthermore, the system analyzes the changing connections between each evolution node to determine how the abnormal state of the target object gradually transitions from one stage to the next. Specifically, the system determines the connection relationships between different nodes based on the time interval between adjacent evolution nodes, the direction of behavioral changes, the magnitude of state changes, and changes in related factors. For example, in a vehicle abnormal parking scenario, if the system detects that a vehicle enters a parking area from a normal driving state, then changes from a short stop to a long stop, and there are changes in human activity around the vehicle, it can be determined that there is a state continuity relationship between the "entry area node" and the "continuous parking node," and a risk enhancement relationship between the "continuous parking node" and the "abnormal human activity node," thus forming a continuous changing connection relationship. Subsequently, based on the changing connections between evolutionary nodes, the system determines the specific development stages in the development of abnormal behavior, dividing the development process of abnormal events into multiple logically related stages. Among them, the behavior triggering stage describes the initial cause or initial change of the abnormal behavior, such as the target object deviating from the normal activity range for the first time; the state transition stage describes the process of the abnormal behavior changing from a normal deviation state to a clearly abnormal state, such as a change in the target object's behavior direction, migration of the activity area, or an enhancement of the interaction relationship; the risk diffusion stage describes the process of the abnormal state further expanding its impact, such as multiple objects participating in the abnormal behavior, expansion of the abnormal area, or an increase in the duration. Finally, based on the relationships between each development stage, the system combines different evolution nodes, state change processes, and development stages to construct an abnormal behavior evolution logic chain. This logic chain describes the complete process of the target object from the generation of an anomaly, state change, to risk expansion according to the development sequence of the abnormal event. For example, for an abnormal wandering event, the system can construct an abnormal behavior evolution logic chain of "entering a sensitive area (behavior trigger) - repeatedly going back and forth and increasing the stay time (state transition) - having abnormal interactions with other objects and continuing to act (risk diffusion)".

[0038] In this embodiment, step S1, which identifies the object type within the target monitoring area based on pre-collected video data from the camera, further includes: S11: Obtain basic information about objects within the target monitoring area, and generate corresponding object recognition results based on the positional change relationship of the target object in different video frames. The basic information about objects specifically includes appearance information, motion state information, and spatial distribution information. S12: Determine whether the object information in the object recognition result meets the preset object association conditions; S13: If so, the target objects are associated and organized to construct a corresponding object relationship description structure. Based on the object relationship description structure, key objects that have an impact on the target monitoring area are selected, and corresponding dynamic object files are established for the key objects.

[0039] In this embodiment, the system acquires basic object information within the target monitoring area. This basic information includes appearance information, motion status information, and spatial distribution information. Based on the positional changes of the target object in different video frames, the system generates corresponding object recognition results. Then, the system determines whether the object information in the object recognition results meets pre-set object association conditions (object association conditions are used to determine whether there is a spatial proximity relationship, motion synchronization relationship, or interactive influence relationship between different objects), and executes corresponding steps accordingly. For example, if the system determines that the object information in the object recognition results cannot meet the pre-set object association conditions, the system will consider it impossible to confirm whether objects appearing at different time points belong to the same target object. The system will analyze the reasons for the object association failure based on the parts of the object recognition results that do not meet the association conditions, extract the influencing factors that cause the objects to fail to match continuously, and supplement and correct the current object information. For example, it may re-extract the appearance features of the target object in adjacent video frames, expand the object position search range, increase the motion trajectory matching window, and re-establish... The system identifies the correspondence between objects across different video frames. For example, when the system determines that the object information in the object recognition result meets the pre-set object association conditions, the system considers that the currently identified object has formed a stable identity association relationship between consecutive video frames. The system will then organize the target object into an associated structure, construct a corresponding object relationship description structure, and based on these object relationship description structures, filter out key objects that have an impact on the target monitoring area and establish a corresponding dynamic object file for the key object. The system can avoid problems such as repeated object recognition and identity confusion caused by unstable single-frame recognition results, temporary occlusion, or environmental changes. This allows subsequent analysis to be carried out based on the complete change process of the same target object, providing a reliable data foundation for continuous behavior analysis and abnormal event identification. At the same time, it can record the appearance, movement state, and spatial location changes of individual objects, and analyze the proximity relationship, interaction relationship, and activity area influence relationship between objects, thereby identifying key objects that play an important role in the formation of abnormal events and improving the accuracy of key target screening in complex multi-target environments.

[0040] It should be noted that the process of associating and organizing the target objects, constructing a corresponding object relationship description structure, and then, based on this structure, selecting key objects that have an impact on the target monitoring area and creating corresponding dynamic object files for these key objects, specifically involves: The system merges data generated by the same object in different time periods based on the appearance features, motion trajectory, spatial position changes, and interactions with other objects of the target object in different video frames, and establishes the correlation between target objects and between target objects and the monitoring area. For example, in the monitoring area of ​​a shopping mall, if the system continuously identifies a person entering from the entrance, passing through the passage, staying in a certain area and contacting another person, it will associate the information of these people identified at different time points to determine that they belong to the same target object and record their complete activity process. Furthermore, based on the associated object information, the system constructs a corresponding object relationship description structure to describe the association status between different objects and the impact of objects on the monitored area. This object relationship description structure can include the object's own state relationship, the interaction relationship between objects, and the positional relationship between objects and spatial areas. Among them, the object's own state relationship is used to describe the changes in the target object's movement speed, activity range, and dwell time; the interaction relationship between objects is used to describe the approach, departure, following, and collaborative behaviors between different objects; and the positional relationship between objects and spatial areas is used to describe whether the target object frequently enters a certain area or changes its activity range. For example, in a parking lot monitoring scenario, if the system finds that a vehicle has been staying near the exit for a long time, and at the same time, people have approached the vehicle multiple times and interacted with it, then the object relationship description structure can be used to identify a strong association between the vehicle, people, and the exit area. Subsequently, based on the object relationship description structure, the system analyzes the degree of influence of each object within the target monitoring area and filters out key objects that have a major impact on changes in the state of the monitoring area. Specifically, the system can evaluate the association of different objects based on factors such as the frequency of an object's participation in abnormal state changes, the number of associations with other objects, the degree of influence of its activity range, and the duration of its behavioral changes, and identify objects that continuously affect changes in the monitoring scene as key objects. For example, in an abnormal event of a crowd gathering, if a person continuously changes the activity area and guides other people to move and gather, then this person has a greater impact on changes in the state of the area than ordinary bystanders, and therefore can be identified as a key object. Finally, the system establishes dynamic object profiles for the selected key objects to continuously record their status changes over different time periods. These profiles include not only basic information about the key objects but also their historical activity trajectories, behavioral patterns, changes in associated objects, and information about their involvement in abnormal events. For example, for personnel identified as key objects, the dynamic object profile can record their location, movement path, area of ​​stay, and interactions with other objects at different times when they enter the monitored area. When similar behavior occurs again, the system can quickly determine whether the object exhibits a recurring trend of abnormal behavior based on the dynamic object profile, improving the continuity and accuracy of abnormal event identification.

[0041] Reference Appendix Figure 2 An abnormal event recognition system for an intelligent security camera, as described in one embodiment of the present invention, includes: The calculation module 10 is used to identify the object type within the target monitoring area based on the video data pre-collected by the camera, generate corresponding target behavior state data according to the target behavior characteristic parameters of the object type in a continuous time period, match and analyze the target behavior state data with a preset set of behavior benchmark features, and calculate the behavior deviation characteristic parameters in the current monitoring scene. The object type specifically includes people, vehicles and objects, and the target behavior characteristic parameters specifically include position change, movement speed, dwell time, activity frequency and interaction relationship. The judgment module 20 is used to determine whether the behavior deviation feature parameter exceeds a preset abnormal deviation threshold; The execution module 30 is used to, if the time exceeds the limit, split the continuous video segments of the video data into behavioral states according to the time order within the continuous time period, obtain the abnormal stage state of the target object, construct the target behavior evolution sequence of the target object based on the abnormal stage state, collect the transition relationship between the abnormal stage states, and obtain the corresponding abnormal behavior evolution path. The abnormal stage state specifically includes the state before the abnormal behavior occurs, the abnormal change process state, and the current abnormal state. The second judgment module 40 is used to judge whether the evolution process of the abnormal behavior evolution path can match the preset abnormal event characteristics. The second execution module 50 is used to mark abnormal candidate objects from the abnormal behavior evolution path if there is no match, extract the regular characteristics of the abnormal candidate objects, generate corresponding abnormal event description features, perform cluster analysis between the abnormal event description features and historical abnormal event data, identify undefined abnormal behaviors with similar evolution trends, classify the undefined abnormal behaviors into preset abnormal event clusters, generate corresponding abnormal labels for the abnormal event clusters, and dynamically expand the abnormal event identification range of the monitoring environment through the abnormal labels. The regular characteristics specifically include behavioral change patterns, spatial activity patterns, and target interaction patterns.

[0042] In this embodiment, the system identifies the object types within the target monitoring area based on video data pre-collected by the camera. Specific object types include people, vehicles, and objects. Based on the target behavior characteristic parameters of different object types over continuous time periods, including position changes, movement speed, dwell time, activity frequency, and interaction relationships, corresponding target behavior state data is generated. This data is then matched and analyzed against a pre-set set of behavioral baseline features (which describes the normal operating state of the monitoring area). Behavioral deviation characteristic parameters in the current monitoring scene are calculated, and the system determines whether these parameters exceed a pre-set abnormal deviation threshold to execute corresponding steps. For example, if the system determines that the behavioral deviation characteristic parameters in the current monitoring scene do not exceed the pre-set abnormal deviation threshold, the system considers the behavior of each object within the current monitoring area to be within a preset normal activity range. The system then updates the normal activity patterns corresponding to different time periods, different spatial areas, and different object types based on the current behavioral state data that does not exceed the abnormal deviation threshold, readjusting the association weights of various target behavior characteristic parameters in the behavioral baseline feature set. For example, if the system determines that the behavioral deviation characteristic parameters in the current monitoring scene do not exceed the pre-set abnormal deviation threshold, the system will proceed accordingly. If the behavioral deviation characteristic parameters exceed a pre-set abnormal deviation threshold, the system will consider the behavioral state of each object within the current monitoring area to be abnormal. The system will then break down the behavioral states of continuous video segments within a consecutive time period, obtaining the abnormal stage states of the target object. These abnormal stage states specifically include the state before the abnormal behavior occurs, the state during the abnormal change process, and the current abnormal state. Based on different abnormal stage states, the system constructs a target behavior evolution sequence for the target object, collects the transition relationships between different abnormal stage states, and obtains the corresponding abnormal behavior evolution path. The system then judges the evolution process of this abnormal behavior evolution path. Can it match the preset abnormal event characteristics to execute the corresponding steps? For example, when the system determines that the evolution process of the abnormal behavior evolution path can match the preset abnormal event characteristics, the system will consider that the current abnormal behavior is not an unknown abnormality, but a known risk event within the preset abnormal event range. The system will extract the key behavior nodes corresponding to the abnormal event based on the matched abnormal event characteristics, including the abnormal behavior trigger time, the abnormal occurrence area, the associated target object, and the abnormal persistence state. Combined with the real-time changes in the current abnormal behavior evolution path, the system will locate the development stage of the abnormal event and send the abnormal event description information to the monitoring and management terminal.For example, when the system determines that the evolution of an abnormal behavior path cannot match preset abnormal event characteristics, the system considers the current abnormal behavior to be an unknown anomaly. The system then marks abnormal candidate objects from the abnormal behavior evolution path, extracts the regular characteristics of these candidate objects (specifically including behavioral change patterns, spatial activity patterns, and target interaction patterns), generates corresponding abnormal event descriptive features, and performs cluster analysis between these features and historical abnormal event data to identify undefined abnormal behaviors with similar evolutionary trends. These undefined abnormal behaviors are then assigned to pre-defined abnormal event clusters, and corresponding abnormal labels are generated for each cluster. These abnormal labels dynamically expand the scope of abnormal event identification in the monitoring environment.

[0043] In this embodiment, it also includes: The generation module is used to perform continuous temporal correlation on the risk video clips based on anomaly deviation analysis, fuse the change information of the target object at different time nodes, and generate the continuous behavioral change trajectory of the target object. The change information specifically includes changes in position, changes in movement direction, changes in behavior duration, and changes in interaction with surrounding objects. The third judgment module is used to determine whether the continuous behavior change trajectory meets the preset continuous evolution conditions; The third execution module is used to identify the changing nodes in different abnormal stages based on the correlation between the abnormal stage states, select the time node with the fastest increase in abnormality from the changing nodes, dynamically divide the time node as the inflection point of the abnormal event evolution, and extract the behavioral combination features not covered by the existing abnormal event categories based on the inflection point of the abnormal event evolution.

[0044] In this embodiment, the system performs continuous temporal correlation on risk video segments based on anomaly deviation analysis ("anomaly deviation analysis" originates from the aforementioned steps of matching real-time acquired target behavior feature parameters with a normal behavior baseline feature set to calculate behavior deviation parameters, and determining risk video segments based on behavior deviation parameters exceeding a preset threshold). It fuses the change information of the target object at different time points, specifically including changes in position, direction of movement, duration of behavior, and interaction with surrounding objects, generating a continuous behavior change trajectory for the target object. The system then determines whether this continuous behavior change trajectory meets preset continuous evolution conditions to execute corresponding steps. For example, if the system determines that the continuous behavior change trajectory of the target object cannot meet the preset continuous evolution conditions, the system considers that although the current target object exhibits... If a certain degree of behavioral deviation occurs, but the behavioral change process does not form a continuous state transition relationship, the system will analyze the influencing factors that cause the interruption of behavioral evolution based on the discontinuous change nodes in the continuous behavioral change trajectory, classify and mark the current behavioral change state, generate corresponding abnormal fluctuation record information, and reduce the abnormal risk level corresponding to the behavioral state to avoid misjudging occasional behavior as a continuous abnormal event. For example, when the system determines that the continuous behavioral change trajectory of the target object can meet the pre-set continuous evolution conditions, the system will consider the abnormal change of the target object to be a frequent state. The system will identify the change nodes in the abnormal stage based on the correlation between different abnormal stage states, select the time node with the fastest increase in abnormality from these change nodes, dynamically divide each time node into abnormal event evolution inflection points, and extract behavioral combination features that are not covered by existing abnormal event categories based on these abnormal event evolution inflection points.

[0045] In this embodiment, the execution module further includes: The generation unit is used to obtain the target object status information corresponding to each abnormal stage based on the abnormal stage status, and to perform association matching on the target object status information according to the temporal relationship between different abnormal stages to generate corresponding stage association information. The target object status information specifically includes the spatial region where the target object is located, the duration of the behavior, the changes in the activity range, and the changes in the environment. The judgment unit is used to determine whether adjacent abnormal stages in the stage association information meet the preset continuous change conditions; The execution unit is configured to, if so, collect state evolution features between adjacent abnormal stages, extract stage transition features of the target object during the behavior change process, identify the behavior evolution features of the target object between different abnormal stages based on the stage transition features, divide the key change nodes when the abnormal state evolves from the initial stage to the subsequent stage, establish a state association structure between abnormal stages through the key change nodes, and combine abnormal stages with continuous evolution relationships to generate a corresponding continuous abnormal state chain. The state evolution features specifically include the state change direction, change magnitude, and change trend, and the behavior evolution features specifically include behavior enhancement, behavior migration, and behavior reversal.

[0046] In this embodiment, the system performs continuous temporal correlation on risk video segments based on anomaly deviation analysis ("anomaly deviation analysis" originates from the aforementioned steps of matching real-time acquired target behavior feature parameters with a normal behavior baseline feature set to calculate behavior deviation parameters, and determining risk video segments based on behavior deviation parameters exceeding a preset threshold). It fuses the change information of the target object at different time points, specifically including changes in position, direction of movement, duration of behavior, and interaction with surrounding objects, generating a continuous behavior change trajectory for the target object. The system then determines whether this continuous behavior change trajectory meets preset continuous evolution conditions to execute corresponding steps. For example, if the system determines that the continuous behavior change trajectory of the target object cannot meet the preset continuous evolution conditions, the system considers that although the current target object exhibits... If a certain degree of behavioral deviation occurs, but the behavioral change process does not form a continuous state transition relationship, the system will analyze the influencing factors that cause the interruption of behavioral evolution based on the discontinuous change nodes in the continuous behavioral change trajectory, classify and mark the current behavioral change state, generate corresponding abnormal fluctuation record information, and reduce the abnormal risk level corresponding to the behavioral state to avoid misjudging occasional behavior as a continuous abnormal event. For example, when the system determines that the continuous behavioral change trajectory of the target object can meet the pre-set continuous evolution conditions, the system will consider the abnormal change of the target object to be a frequent state. The system will identify the change nodes in the abnormal stage based on the correlation between different abnormal stage states, select the time node with the fastest increase in abnormality from these change nodes, dynamically divide each time node into abnormal event evolution inflection points, and extract behavioral combination features that are not covered by existing abnormal event categories based on these abnormal event evolution inflection points.

[0047] In this embodiment, the second execution module further includes: The second generation unit is used to obtain the set of change information corresponding to each abnormal stage based on the abnormal behavior evolution path, perform correlation analysis on different state nodes in the abnormal behavior evolution path, construct the change relationship between each state node, and filter target objects with continuous change characteristics according to the change relationship to generate a corresponding set of candidate objects. The set of change information specifically includes target object state change information, spatial activity change information, and related object change information. The second judgment unit is used to judge whether the target object in the candidate object set satisfies the preset abnormal evolution association condition; The second execution unit is configured to, if so, acquire the behavioral performance information of the target object in different evolutionary stages, collect the change patterns of the target object in different dimensions based on the behavioral performance information, construct the behavioral description structure of the target object based on the correlation between the change patterns, and classify the change combination features of existing abnormal categories through the behavioral description structure, wherein the different dimensions specifically include time dimension, space dimension and interaction dimension.

[0048] In this embodiment, based on the abnormal behavior evolution path, the system obtains a set of change information corresponding to each abnormal stage. Specifically, the set of change information includes target object state change information, spatial activity change information, and related object change information. The system performs correlation analysis on different state nodes in the abnormal behavior evolution path, constructs change relationships between each state node, and filters target objects with continuous change characteristics based on different change relationships, generating a corresponding candidate object set. Then, the system determines whether the target objects in the candidate object set meet the pre-set abnormal evolution correlation conditions (the abnormal evolution correlation conditions are used to determine whether the target object continuously participates in abnormal state changes during the development of abnormal behavior, including the number of abnormal stages the target object participates in, the degree of continuity of state changes, and the correlation changes with other objects), and executes the corresponding steps accordingly. For example, when the system determines that the target objects in the candidate object set cannot meet the pre-set abnormal evolution correlation conditions, the system will consider that although the current candidate object appears in the abnormal behavior evolution path, it has not continuously participated in the abnormal state change. During the development of the state, the system analyzes the specific reasons why the target object fails to meet the abnormal evolution association conditions based on its participation in the abnormal behavior evolution path. It then re-screens the target objects in the candidate object set, marking objects that cannot continuously influence the abnormal state change as low-association objects and removing them from subsequent abnormal event evolution analysis to avoid bias in abnormal event identification results due to the participation of irrelevant objects. For example, when the system determines that a target object in the candidate object set meets the pre-set abnormal evolution association conditions, it considers the target object to have a high degree of association with the abnormal event and requires it to be the subject of abnormal event analysis. The system acquires the target object's behavioral performance information at different evolution stages and collects the target object's change patterns in different dimensions, including time, space, and interaction. Based on the correlation between these change patterns, it constructs the target object's behavioral description structure and uses this structure to classify the change combination features of existing abnormal categories.

[0049] In this embodiment, the determination module further includes: A unit is established to establish a corresponding deviation association structure based on the behavioral deviation feature parameters; The third judgment unit is used to determine whether the deviation factors in the deviation association structure meet the preset association evolution conditions; The third execution unit is used to, if so, obtain the main influencing factors in the target object's behavior change process according to the deviation association structure, collect the evolution order among the main influencing factors, divide the abnormal changes generated by the target object in a continuous time period into stages according to the evolution order, and identify the deviation process caused by behavior change. Specifically, the deviation process includes the local deviation process caused by a single behavior change and the composite deviation process formed by the combined action of multiple behavior factors.

[0050] In this embodiment, the system establishes a corresponding deviation correlation structure based on these behavioral deviation feature parameters (the deviation correlation structure is used to describe the mutual influence relationship between various behavioral factors during the abnormal change of the target object). Then, the system determines whether the deviation factors in the deviation correlation structure meet the pre-set correlation evolution conditions (the correlation evolution conditions are used to determine whether there is a synchronous change, continuous change, or mutual influence relationship between different deviation factors), and executes the corresponding steps accordingly. For example, when the system determines that the deviation factors in the deviation correlation structure cannot meet the pre-set correlation evolution conditions, the system considers that although each deviation factor may exhibit a certain degree of abnormal fluctuation, it has not formed a mutually influential, co-evolving development process. The system will then, based on the degree of correlation between the deviation factors in the deviation correlation structure, split the deviation factors that cannot form a correlation evolution relationship and identify them. Each deviation factor corresponds to an independent source of change. Each independent deviation factor is analyzed separately, and its occurrence time, area of ​​effect, magnitude of change, and duration are extracted, generating corresponding deviation status record information. For example, when the system determines that the deviation factors in the deviation correlation structure can meet the pre-set correlation evolution conditions, the system will consider that there is a continuous correlation change relationship between the behavioral deviation factors in the current monitoring scenario. Based on the deviation correlation structure, the system will obtain the main influencing factors in the behavioral change process of the target object, collect the evolution sequence between these main influencing factors, and divide the abnormal changes generated by the target object in a continuous time according to different evolution sequences. The system will identify the deviation process caused by behavioral changes. The deviation process specifically includes the local deviation process caused by a single behavioral change and the composite deviation process formed by the combined action of multiple behavioral factors.

[0051] In this embodiment, the second determination module further includes: The third generation unit is used to obtain the correlation strength between each evolution stage based on the abnormal behavior evolution path, and to analyze the evolution structure in the abnormal behavior evolution path according to the correlation strength, so as to generate path structure parameters of the abnormal behavior development law. The fourth judgment unit is used to determine whether the path structure parameters meet the preset path stability conditions; The fourth execution unit is used to extract the evolution nodes of the abnormal behavior of the target object from the initial state to the subsequent state according to the stage association relationship in the abnormal behavior evolution path, identify the change connection relationship between the evolution nodes, determine the development links in the abnormal behavior development process, and construct the abnormal behavior evolution logic chain. The development links specifically include the behavior triggering link, the state transition link, and the risk diffusion link.

[0052] In this embodiment, the system obtains the correlation strength between each evolution stage based on the abnormal behavior evolution path. According to different correlation strengths, it analyzes the evolutionary structure in the abnormal behavior evolution path to generate path structure parameters of the abnormal behavior development pattern. Then, the system determines whether these path structure parameters meet pre-set path stability conditions (path stability conditions are used to determine whether there is a continuous state change relationship and a clear development direction during the abnormal behavior evolution process), and executes the corresponding steps accordingly. For example, when the system determines that the path structure parameters of the abnormal behavior development pattern cannot meet the pre-set path stability conditions, the system will consider that the abnormal state of the target object may have stage jumps, interruptions in change, or random fluctuations, and cannot form a stable abnormal development trend. The system will analyze the abnormal behavior evolution path based on the parts of the path structure parameters that do not meet the stability conditions to identify the causes of path instability. Based on certain influencing factors, the unstable stages in the evolution path of abnormal behavior are reclassified, and abnormal stages that cannot form a continuous evolutionary relationship are separated. The time information, spatial information, and behavioral change information corresponding to each stage are retained to generate independent abnormal state records. For example, when the system determines that the path structure parameters of the abnormal behavior development law can meet the pre-set path stability conditions, the system will consider that there is a correlation between the evolution stages in the current abnormal behavior evolution path, which can form a stable abnormal development trend. Based on the stage correlation relationship in the abnormal behavior evolution path, the system will extract the evolution nodes in the process of the target object's abnormal behavior changing from the initial state to the subsequent state, identify the change connection relationship between these evolution nodes, determine the development links in the abnormal behavior development process, and the development links specifically include the behavior triggering link, the state transition link, and the risk diffusion link, thus constructing an abnormal behavior evolution logic chain.

[0053] In this embodiment, the calculation module further includes: The fourth generation unit is used to acquire basic information about objects within the target monitoring area, and generate corresponding object recognition results based on the positional change relationship of the target object in different video frames. Specifically, the basic information about the object includes appearance information, motion state information, and spatial distribution information. The fifth judgment unit is used to determine whether the object information in the object recognition result meets the preset object association conditions; The fifth execution unit is used to, if so, associate and organize the target object, construct a corresponding object relationship description structure, filter key objects that have an impact on the target monitoring area according to the object relationship description structure, and establish corresponding dynamic object files for the key objects.

[0054] In this embodiment, the system acquires basic object information within the target monitoring area. This basic object information specifically includes appearance information, motion state information, and spatial distribution information. Based on the positional changes of the target object in different video frames, a corresponding object recognition result is generated. The system then determines whether the object information in the object recognition result meets pre-set object association conditions (object association conditions are used to determine whether there is a spatial proximity relationship, motion synchronization relationship, or interactive influence relationship between different objects), and executes corresponding steps accordingly. For example, if the system determines that the object information in the object recognition result does not meet the pre-set object association conditions, the system will consider it impossible to confirm whether objects appearing at different time points belong to the same target object, and the system will proceed according to the object recognition result that does not meet the association conditions. In this section, the reasons for object association failure are analyzed, the influencing factors that prevent continuous object matching are extracted, and the current object information is supplemented and corrected. For example, the appearance features of the target object in adjacent video frames are re-extracted, the object location search range is expanded, the motion trajectory matching window is added, and the object correspondence between different video frames is re-established. For example, when the system determines that the object information in the object recognition result can meet the pre-set object association conditions, the system will consider that the currently recognized object has formed a stable identity association relationship between consecutive video frames. The system will then organize the target object into associations, construct the corresponding object relationship description structure, and based on these object relationship description structures, select key objects that have an impact on the target monitoring area and establish corresponding dynamic object files for these key objects.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying abnormal events in an intelligent security camera, characterized in that, Includes the following steps: Based on video data pre-collected by the camera in the target monitoring area, the object type in the target monitoring area is identified. According to the target behavior feature parameters of the object type in a continuous time period, corresponding target behavior state data is generated. The target behavior state data is matched and analyzed with a preset set of behavior benchmark features to calculate the behavior deviation feature parameters in the current monitoring scene. The object type specifically includes people, vehicles and objects. The target behavior feature parameters specifically include position change, movement speed, dwell time, activity frequency and interaction relationship. Determine whether the behavioral deviation feature parameter exceeds a preset abnormal deviation threshold; If the time exceeds the limit, the video data is split into behavioral states in the continuous video segments according to the time order within the continuous time period to obtain the abnormal stage state of the target object. Based on the abnormal stage state, the target behavior evolution sequence of the target object is constructed, the transition relationship between the abnormal stage states is collected, and the corresponding abnormal behavior evolution path is obtained. The abnormal stage state specifically includes the state before the abnormal behavior occurs, the abnormal change process state, and the current abnormal state. Determine whether the evolution process of the abnormal behavior evolution path matches the preset abnormal event characteristics; If there is no match, abnormal candidate objects are marked from the abnormal behavior evolution path, the regular characteristics of the abnormal candidate objects are extracted, corresponding abnormal event description features are generated, and cluster analysis is performed between the abnormal event description features and historical abnormal event data to identify undefined abnormal behaviors with similar evolution trends. The undefined abnormal behaviors are divided into preset abnormal event clusters, and corresponding abnormal labels are generated for the abnormal event clusters. Through the abnormal labels, the abnormal event identification range of the monitoring environment is dynamically expanded. The regular characteristics specifically include behavioral change patterns, spatial activity patterns, and target interaction patterns.

2. The abnormal event recognition method for intelligent security cameras according to claim 1, characterized in that, After the step of splitting the continuous video segments of the video data into behavioral states according to the time sequence within a continuous time period to obtain the abnormal stage state of the target object, the method further includes: Based on the risk video clips of anomaly deviation analysis, the target objects in the risk video clips are continuously correlated over time, and the change information of the target objects at different time nodes is fused to generate the continuous behavioral change trajectory of the target objects. The change information specifically includes changes in position, changes in movement direction, changes in behavior duration, and changes in interaction with surrounding objects. Determine whether the continuous behavior change trajectory meets the preset continuous evolution conditions; If so, based on the correlation between the abnormal stage states, identify the changing nodes in different abnormal stages, select the time node with the fastest increase in abnormality from the changing nodes, dynamically divide the time node as the inflection point of the abnormal event evolution, and extract the behavioral combination features that are not covered by the existing abnormal event categories based on the inflection point of the abnormal event evolution.

3. The abnormal event recognition method for intelligent security cameras according to claim 1, characterized in that, The step of constructing the target behavior evolution sequence of the target object based on the abnormal stage state, collecting the transition relationship between the abnormal stage states, and obtaining the corresponding abnormal behavior evolution path further includes: Based on the abnormal stage state, the target object state information corresponding to each abnormal stage is obtained. According to the time sequence relationship between different abnormal stages, the target object state information is associated and matched to generate corresponding stage association information. The target object state information specifically includes the spatial area where the target object is located, the duration of the behavior, the changes in the activity range, and the changes in the environment. Determine whether adjacent abnormal stages in the stage association information meet the preset continuous change condition; If so, then the state evolution features between adjacent abnormal stages are collected, the stage transition features of the target object in the process of behavioral change are extracted, and based on the stage transition features, the behavioral evolution features of the target object between different abnormal stages are identified. The key change nodes when the abnormal state evolves from the initial stage to the subsequent stage are divided. Through the key change nodes, the state association structure between abnormal stages is established, and abnormal stages with continuous evolution relationship are combined to generate the corresponding continuous abnormal state chain. The state evolution features specifically include the state change direction, change magnitude and change trend, and the behavioral evolution features specifically include behavior enhancement, behavior migration and behavior reversal.

4. The abnormal event recognition method for intelligent security cameras according to claim 1, characterized in that, The step of marking abnormal candidate objects from the abnormal behavior evolution path, extracting the regular characteristics of the abnormal candidate objects, and generating corresponding abnormal event description features further includes: Based on the abnormal behavior evolution path, the change information set corresponding to each abnormal stage is obtained, the correlation analysis of different state nodes in the abnormal behavior evolution path is performed, the change relationship between each state node is constructed, and the target objects with continuous change characteristics are screened according to the change relationship to generate the corresponding candidate object set. The change information set specifically includes target object state change information, spatial activity change information, and related object change information. Determine whether the target object in the candidate object set satisfies the preset abnormal evolution association condition; If so, then obtain the behavioral performance information of the target object in different evolutionary stages, collect the change patterns of the target object in different dimensions based on the behavioral performance information, construct the behavioral description structure of the target object based on the correlation between the change patterns, and divide the change combination features of existing abnormal categories through the behavioral description structure. The different dimensions specifically include time dimension, space dimension and interaction dimension.

5. The abnormal event recognition method for intelligent security cameras according to claim 1, characterized in that, The step of determining whether the behavioral deviation feature parameter exceeds a preset abnormal deviation threshold further includes: Based on the aforementioned behavioral deviation feature parameters, a corresponding deviation association structure is established; Determine whether the deviation factors in the deviation association structure meet the preset association evolution conditions; If so, based on the deviation association structure, the main influencing factors in the target object's behavior change process are obtained, the evolution order among the main influencing factors is collected, and based on the evolution order, the abnormal changes generated by the target object in a continuous time period are divided into stages to identify the deviation process caused by behavior change. Specifically, the deviation process includes the local deviation process caused by a single behavior change and the composite deviation process formed by the combined action of multiple behavior factors.

6. The abnormal event recognition method for intelligent security cameras according to claim 1, characterized in that, The step of determining whether the evolution process of the abnormal behavior evolution path can match the preset abnormal event characteristics further includes: Based on the abnormal behavior evolution path, the correlation strength between each evolution stage is obtained. According to the correlation strength, the evolution structure in the abnormal behavior evolution path is analyzed to generate path structure parameters of the abnormal behavior development law. Determine whether the path structure parameters meet the preset path stability conditions; If so, based on the stage association relationship in the abnormal behavior evolution path, the evolution nodes in the process of the abnormal behavior of the target object changing from the initial state to the subsequent state are extracted, the change connection relationship between the evolution nodes is identified, the development links in the abnormal behavior development process are determined, and the abnormal behavior evolution logic chain is constructed. The development links specifically include the behavior triggering link, the state transition link, and the risk diffusion link.

7. The abnormal event recognition method for intelligent security cameras according to claim 1, characterized in that, The step of identifying the object type within the target monitoring area based on pre-collected video data from a camera on the target monitoring area further includes: The basic information of objects within the target monitoring area is obtained, and the corresponding object recognition result is generated based on the positional change relationship of the target object in different video frames. The basic information of the object specifically includes appearance information, motion state information and spatial distribution information. Determine whether the object information in the object recognition result meets the preset object association conditions; If so, the target objects are associated and organized to construct a corresponding object relationship description structure. Based on the object relationship description structure, key objects that have an impact on the target monitoring area are selected, and corresponding dynamic object files are established for the key objects.

8. An abnormal event recognition system for an intelligent security camera, characterized in that, include: The calculation module is used to identify the object type within the target monitoring area based on the video data pre-collected by the camera, generate corresponding target behavior state data according to the target behavior feature parameters of the object type in a continuous time period, match and analyze the target behavior state data with a preset set of behavior benchmark features, and calculate the behavior deviation feature parameters in the current monitoring scene. The object type specifically includes people, vehicles and objects, and the target behavior feature parameters specifically include position change, movement speed, dwell time, activity frequency and interaction relationship. The judgment module is used to determine whether the behavior deviation feature parameter exceeds a preset abnormal deviation threshold; The execution module is used to, if the time exceeds the specified time period, split the continuous video segments of the video data into behavioral states according to the time order within the continuous time period, obtain the abnormal stage state of the target object, construct the target behavior evolution sequence of the target object based on the abnormal stage state, collect the transition relationship between the abnormal stage states, and obtain the corresponding abnormal behavior evolution path. The abnormal stage state specifically includes the state before the abnormal behavior occurs, the abnormal change process state, and the current abnormal state. The second judgment module is used to determine whether the evolution process of the abnormal behavior evolution path can match the preset abnormal event characteristics. The second execution module is used to mark abnormal candidate objects from the abnormal behavior evolution path if there is no match, extract the regular characteristics of the abnormal candidate objects, generate corresponding abnormal event description features, perform cluster analysis between the abnormal event description features and historical abnormal event data, identify undefined abnormal behaviors with similar evolution trends, classify the undefined abnormal behaviors into preset abnormal event clusters, generate corresponding abnormal labels for the abnormal event clusters, and dynamically expand the abnormal event identification range of the monitoring environment through the abnormal labels. The regular characteristics specifically include behavioral change patterns, spatial activity patterns, and target interaction patterns.

9. The abnormal event recognition system for intelligent security cameras according to claim 8, characterized in that, Also includes: The generation module is used to perform continuous temporal correlation on the risk video clips based on anomaly deviation analysis, fuse the change information of the target object at different time nodes, and generate the continuous behavioral change trajectory of the target object. The change information specifically includes changes in position, changes in movement direction, changes in behavior duration, and changes in interaction with surrounding objects. The third judgment module is used to determine whether the continuous behavior change trajectory meets the preset continuous evolution conditions; The third execution module is used to identify the changing nodes in different abnormal stages based on the correlation between the abnormal stage states, select the time node with the fastest increase in abnormality from the changing nodes, dynamically divide the time node as the inflection point of the abnormal event evolution, and extract the behavioral combination features not covered by the existing abnormal event categories based on the inflection point of the abnormal event evolution.

10. The abnormal event recognition system for intelligent security cameras according to claim 8, characterized in that, The execution module further includes: The generation unit is used to obtain the target object status information corresponding to each abnormal stage based on the abnormal stage status, and to perform association matching on the target object status information according to the temporal relationship between different abnormal stages to generate corresponding stage association information. The target object status information specifically includes the spatial region where the target object is located, the duration of the behavior, the changes in the activity range, and the changes in the environment. The judgment unit is used to determine whether adjacent abnormal stages in the stage association information meet the preset continuous change conditions; The execution unit is configured to, if so, collect state evolution features between adjacent abnormal stages, extract stage transition features of the target object during the behavior change process, identify the behavior evolution features of the target object between different abnormal stages based on the stage transition features, divide the key change nodes when the abnormal state evolves from the initial stage to the subsequent stage, establish a state association structure between abnormal stages through the key change nodes, and combine abnormal stages with continuous evolution relationships to generate a corresponding continuous abnormal state chain. The state evolution features specifically include the state change direction, change magnitude, and change trend, and the behavior evolution features specifically include behavior enhancement, behavior migration, and behavior reversal.

Images