Scene-adaptive anomaly monitoring method and device and storage medium

Abstract

The invention relates to the technical field of computers, in particular to a scene-adaptive anomaly monitoring method and device and a storage medium, and the method comprises the steps: obtaining a multi-modal environment parameter in a monitoring region, and generating a current scene feature vector according to the multi-modal environment parameter; determining a target monitoring algorithm from a preset algorithm strategy library based on the current scene feature vector; obtaining video data collected by a camera device, and determining target behavior feature data according to the target monitoring algorithm and the video data; and determining an anomaly monitoring strategy according to the target behavior feature data. The method and the device have the effect of improving the accuracy and the processing efficiency of anomaly monitoring in a complex scene.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a scene-adaptive anomaly monitoring method, device, and storage medium. Background Technology

[0002] Currently, video surveillance systems have been widely used in various fields such as urban security, traffic management, and industrial production, becoming an important technical means to maintain public safety and improve management efficiency.

[0003] Existing video surveillance systems typically rely on pre-defined, single, or fixed algorithm models when detecting abnormal behavior. For example, a system might be configured with a pedestrian detection and tracking algorithm for typical daytime scenarios. However, when the actual monitoring scenario changes—such as from daytime to nighttime, from sunny to rainy, or from sparse to crowded population density—this fixed algorithm model often struggles to adapt, leading to a sharp decline in monitoring performance. Specifically, this manifests as an increased false negative rate for genuine abnormal events or an increased false positive rate due to environmental interference, thus indicating room for improvement. Summary of the Invention

[0004] To improve the accuracy and efficiency of anomaly monitoring in complex scenarios, this application provides a scenario-adaptive anomaly monitoring method, device, and storage medium.

[0005] The above-mentioned objective of this application is achieved through the following technical solution: A scene-adaptive anomaly detection method, the anomaly detection method comprising: Acquire multimodal environmental parameters within the monitoring area, and generate a current scene feature vector based on the multimodal environmental parameters; Based on the current scene feature vector, a target detection algorithm is determined from a preset algorithm strategy library; Acquire video data collected by the camera device, and determine target behavior feature data based on the target monitoring algorithm and the video data; Based on the target behavior characteristic data, an anomaly monitoring strategy is determined.

[0006] By adopting the above technical solution, and by acquiring multimodal environmental parameters and generating scene feature vectors, the environment can be quantified and accurately perceived, thus providing an objective basis for subsequent decision-making. By dynamically determining the most suitable target monitoring algorithm based on the scene feature vectors, it can be ensured that video analysis is always performed under the optimal algorithm model, which greatly improves the accuracy and robustness of target detection and tracking in complex scenarios such as changes in lighting and personnel density. Furthermore, based on the accurately extracted target behavior feature data, anomaly monitoring strategies are determined, realizing a fully intelligent closed loop from environmental perception to algorithm selection, and then to behavior analysis and strategy response, effectively improving the overall performance of anomaly monitoring.

[0007] In a preferred embodiment, this application can be further configured such that: generating the current scene feature vector based on the multimodal environment parameters specifically includes: The multimodal environmental parameters within the monitoring area are collected using optical sensors, infrared sensors, and a microphone array. These multimodal environmental parameters include at least light intensity data, environmental thermal distribution data, and background noise data. Based on the multimodal environment parameters, the current scene feature vector is determined through a preset multilayer perception model.

[0008] By adopting the above technical solution and integrating environmental parameters from multiple dimensions such as illumination, heat, and noise, a more comprehensive and three-dimensional scene description can be constructed than that of single visual information. This effectively overcomes the perception limitations of traditional visual solutions in nighttime, inclement weather, or under occlusion. By utilizing a multi-layer perception model to deeply fuse these heterogeneous data, deeper scene association features can be mined, thereby generating more accurate and robust scene feature vectors, providing high-quality input for the adaptive selection of subsequent algorithms.

[0009] In a preferred embodiment, this application can be further configured as follows: determining the target detection algorithm from a preset algorithm strategy library based on the current scene feature vector specifically includes: The current scene feature vector is matched with the algorithm strategy library to obtain the matching degree result of the monitoring algorithm; When the highest matching score in the matching results is greater than the preset matching threshold, the preset monitoring algorithm that best matches the algorithm strategy library is determined as the target monitoring algorithm.

[0010] By adopting the above technical solution and establishing a quantitative matching degree and threshold judgment mechanism, a clear, objective and automated decision-making standard is provided for algorithm selection, avoiding the blindness of manual intervention or the use of fixed strategies. This mechanism ensures that the system can always select an algorithm that is highly adapted to the current environment, thereby maintaining high monitoring performance in various dynamically changing scenarios and enhancing the intelligence and reliability of the entire system.

[0011] In a preferred embodiment, this application can be further configured as follows: after determining the target monitoring algorithm, the currently running algorithm is unloaded, and the target monitoring algorithm is loaded and run.

[0012] By adopting the above technical solution, the on-demand allocation and efficient utilization of computing resources are achieved by dynamically unloading old algorithms and loading new ones. This mechanism ensures that hardware resources always serve the current optimal monitoring task, avoiding resource waste caused by running unsuitable algorithms. Especially in resource-constrained deployment environments such as edge computing, it can significantly improve the system's operating efficiency and response speed.

[0013] In a preferred embodiment, this application can be further configured such that: determining the target behavior feature data based on the target detection algorithm and the video data specifically includes: The target detection algorithm is used to analyze the video data to determine the target objects in the video data; The target object is associated with the video data to generate motion trajectory data of the target object. Based on the motion trajectory data, the target behavior feature data of the target object is determined. The target behavior feature data includes at least the movement speed and the duration of stay in the monitoring area.

[0014] By adopting the above technical solution, through detection followed by tracking, it is possible to accurately separate each independent target from the video stream and establish a unique identity and continuous motion trajectory for it. The trajectory data is quantified into specific behavioral characteristics such as movement speed and dwell time, which transforms the description of target behavior from post-event manual qualitative judgment to real-time machine quantitative analysis, providing an objective and structured data foundation for the accurate identification of subsequent abnormal behaviors.

[0015] In a preferred embodiment, this application can be further configured such that: determining the anomaly monitoring strategy based on the target behavior feature data specifically includes: Based on the target behavior feature data, a spatiotemporal behavior map is constructed; The spatiotemporal behavior map is input into a preset abnormal behavior classification model for analysis, and abnormal results are obtained from the abnormal behavior classification model. Based on the abnormal results, the abnormality monitoring strategy is determined.

[0016] By adopting the above technical solutions and constructing a spatiotemporal behavior map, isolated target behavior feature data can be integrated into structured information that reflects the complex interaction between targets and the environment, and between targets themselves. By using an abnormal behavior classification model to analyze the map, the overall situation within the scene can be understood from a higher dimension and a broader perspective. This enables the identification of complex abnormal patterns, such as multi-person collaboration and abnormal clustering, which cannot be detected by analyzing a single target alone, thus greatly improving the depth and breadth of anomaly monitoring.

[0017] In a preferred embodiment, this application can be further configured such that: the construction of a spatiotemporal behavior map based on the target behavior feature data specifically includes: The monitoring area is divided into multiple area nodes, and each target object in the video data is identified as a target node; The motion trajectory data of the target object is spatially mapped with multiple region nodes to generate a region node sequence and trajectory data. Based on the trajectory data, association edges are constructed between the target node and the regional node, as well as between different regional nodes, to obtain the spatiotemporal behavior graph.

[0018] By adopting the above technical solution, a specific and feasible path for constructing a spatiotemporal behavior graph is provided. By defining region nodes and target nodes, static and dynamic elements in the scene are clearly distinguished. Through spatial mapping and the construction of associated edges, the spatiotemporal behavior of the target is effectively transformed into the relationship between nodes and edges in graph theory, so that complex real-world dynamics can be understood and processed efficiently, and high-quality, structured input data is provided for subsequent graph analysis algorithms.

[0019] In a preferred embodiment, this application can be further configured such that: determining the anomaly monitoring strategy based on the anomaly result specifically includes: If the confidence level in the abnormal result is less than the first threshold, then the abnormal monitoring strategy is determined to be a tracking and recording strategy. If the confidence level in the abnormal result is greater than the first threshold and less than the second threshold, then the abnormal monitoring strategy is determined to be a push alarm strategy. If the confidence level of the abnormal result is greater than the second threshold, then the abnormal monitoring strategy is determined to be an on-site intervention strategy.

[0020] By adopting the above technical solution and setting different threshold ranges according to the confidence level of abnormal results, differentiated and graded responses to events of different risk levels can be achieved. This avoids overreaction to low-confidence events, saves the energy of security personnel, and ensures timely attention to medium- and high-risk events and rapid intervention in high-risk events. This makes the allocation of security resources more precise and reasonable, and greatly improves the efficiency of handling abnormal events and the closed-loop management capability.

[0021] The above-mentioned objective 2 of this application is achieved through the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described scene-adaptive anomaly detection method.

[0022] The above-mentioned objective three of this application is achieved through the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described scenario-adaptive anomaly monitoring method.

[0023] In summary, this application includes at least one of the following beneficial technical effects: 1. By acquiring multimodal environmental parameters and generating scene feature vectors, the environment can be quantified and accurately perceived, thus providing an objective basis for subsequent decision-making. By dynamically determining the most suitable target monitoring algorithm based on the scene feature vectors, it can ensure that video analysis is always performed under the optimal algorithm model, greatly improving the accuracy and robustness of target detection and tracking in complex scenarios such as changes in lighting and personnel density. Furthermore, based on the accurately extracted target behavior feature data, anomaly monitoring strategies are determined, realizing a fully intelligent closed loop from environmental perception to algorithm selection, behavior analysis, and strategy response, effectively improving the overall performance of anomaly monitoring.

[0024] 2. By dynamically unloading old algorithms and loading new ones, on-demand allocation and efficient utilization of computing resources are achieved. This mechanism ensures that hardware resources are always used for the current optimal monitoring task, avoiding resource waste caused by running unsuitable algorithms. Especially in resource-constrained deployment environments such as edge computing, it can significantly improve the system's operating efficiency and response speed.

[0025] 3. By constructing a spatiotemporal behavior map, isolated target behavior feature data can be integrated into structured information that reflects the complex interaction between the target and the environment, and between targets. By using an abnormal behavior classification model to analyze the map, the overall situation within the scene can be understood from a higher dimension and a broader perspective. This enables the identification of complex abnormal patterns, such as multi-person collaboration and abnormal clustering, which cannot be discovered by analyzing a single target alone, thus greatly improving the depth and breadth of anomaly monitoring. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the implementation of a scene-adaptive anomaly detection method in one embodiment of this application. Figure 2 This is a flowchart illustrating the implementation of step S10 in a scene-adaptive anomaly detection method according to an embodiment of this application. Figure 3 This is a flowchart illustrating the implementation of step S20 in a scene-adaptive anomaly detection method according to an embodiment of this application. Figure 4 This is a flowchart illustrating the implementation of step S30 in a scene-adaptive anomaly detection method according to an embodiment of this application. Figure 5 This is a flowchart illustrating the implementation of step S40 in a scene-adaptive anomaly detection method according to an embodiment of this application. Figure 6 This is a flowchart illustrating the implementation of step S42 in a scene-adaptive anomaly detection method according to an embodiment of this application. Figure 7 This is a flowchart illustrating the implementation of step S43 in a scene-adaptive anomaly detection method according to an embodiment of this application. Figure 8 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application. Detailed Implementation

[0027] The following embodiments will help those skilled in the art to further understand the function of this application, but do not limit this application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application. These all fall within the protection scope of this application.

[0028] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0029] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0030] The present application will be further described in detail below with reference to the accompanying drawings.

[0031] In one embodiment, such as Figure 1 As shown, this application discloses a scene-adaptive anomaly detection method, which specifically includes the following steps: S10: Obtain multimodal environmental parameters within the monitoring area, and generate the current scene feature vector based on the multimodal environmental parameters.

[0032] Specifically, in order to provide a digital and structured description of the monitoring environment, it is necessary to collect and integrate raw data from different types of sensors. These data reflect multiple physical dimensions of the environment. After being processed by a preset feature extraction model, they can be transformed into a numerical vector that can be understood and compared by a computer. This vector is like a digital fingerprint of the scene, uniquely identifying the overall state of the current environment.

[0033] S20: Based on the current scene feature vector, determine the target detection algorithm from the preset algorithm strategy library.

[0034] Specifically, the scene feature vector generated in the previous step is used as a query condition to perform a search and matching in a pre-built algorithm library. This library contains a variety of video analysis algorithms, each of which is labeled with one or more scene features that it is most suitable for. By calculating the similarity between the current scene feature vector and the scene features labeled by each algorithm in the library, the algorithm that best fits the current environment can be found, thereby realizing the intelligent selection of algorithms.

[0035] S30: Acquire video data collected by the camera equipment, and determine the target behavior feature data based on the target monitoring algorithm and the video data.

[0036] Specifically, the target monitoring algorithm selected in the previous step is invoked to analyze and process the real-time video data input from the camera device. This algorithm can identify moving targets from continuous video frames and continuously track them, thereby recording the motion process of each target and calculating a series of key quantitative indicators from these processes, such as the target's speed, direction of travel, and frequency of appearance in certain areas. These indicators together constitute feature data describing its behavior pattern.

[0037] S40: Determine the anomaly monitoring strategy based on the target behavior characteristic data.

[0038] Specifically, the extracted target behavioral characteristic data undergoes in-depth analysis, going beyond simply evaluating the isolated behavior of individual targets. It focuses on analyzing the spatiotemporal interactions between targets and their environment, and between targets themselves, and then inputs these relationships into a pre-defined risk assessment model for reasoning. This model outputs a quantitative assessment of the degree of current behavioral abnormality. Based on this assessment, it automatically selects the most appropriate strategy from multiple pre-defined strategies of varying response intensities to execute, thereby achieving precise and appropriate handling of abnormal situations.

[0039] In one embodiment, such as Figure 2 As shown, in step S10, the current scene feature vector is generated based on the multimodal environment parameters, which specifically includes: S11: Collect multimodal environmental parameters within the monitoring area using optical sensors, infrared sensors, and microphone arrays. These multimodal environmental parameters include at least light intensity data, environmental thermal distribution data, and background noise data.

[0040] Specifically, in order to comprehensively perceive the environment, multiple sensors can be deployed to work together. For example, optical sensors can be used to measure the lux value of the current illumination to determine whether it is daytime, dusk or nighttime. Infrared sensor arrays can be used to capture the distribution of thermal radiation in space to detect whether there are abnormal heat sources or the gathering of organisms in the environment. At the same time, microphone arrays can be used to collect the decibel value and spectral characteristics of background sound to distinguish whether the environment is quiet, noisy or has sudden abnormal noises.

[0041] S12: Based on multimodal environment parameters, the current scene feature vector is determined through a preset multi-layer perception model.

[0042] Specifically, the collected data, such as light intensity values, heat map data, and background noise features, which are of different formats and dimensions, are normalized and then input as a whole into a pre-trained multilayer perceptron neural network model. Through its internal multilayer nonlinear transformations, the model can effectively fuse these heterogeneous information and learn deep features that can represent the essential state of the scene. Finally, it outputs a fixed-dimensional floating-point vector as the feature vector of the current scene.

[0043] In one embodiment, such as Figure 3 As shown, in step S20, which involves determining a target detection algorithm from a preset algorithm strategy library based on the current scene feature vector, the specific steps include: S21: Match the current scene feature vector with the algorithm strategy library to obtain the matching degree result of the monitoring algorithm.

[0044] Specifically, the system pre-stores an algorithm strategy library containing various monitoring algorithms optimized for different scenarios. For example, there are image enhancement combined with target detection algorithms suitable for low-light environments at night, and cross-camera tracking algorithms suitable for dense crowd scenes. Each algorithm is associated with one or more most suitable scene feature vectors. The generated current scene feature vector is compared with the scene feature vectors associated with each algorithm in the algorithm strategy library to calculate similarity, for example, by using a cosine similarity algorithm, to obtain a series of matching scores between 0 and 1. This score intuitively reflects the degree of adaptability of each algorithm to the current environment, thus obtaining a series of matching results.

[0045] S22: When the highest matching score in the matching results is greater than the preset matching threshold, the preset monitoring algorithm that matches best in the algorithm strategy library will be determined as the target monitoring algorithm.

[0046] Specifically, to avoid making incorrect choices when no suitable algorithm is available, a matching threshold of, for example, 0.85 can be set. After obtaining the matching scores of all algorithms, the highest score is found first. If the highest score, for example, 0.92, is greater than the set threshold of 0.85, it indicates that an algorithm that is highly suitable for the current scene has been found. At this point, the algorithm with the highest score is determined as the target monitoring algorithm to be used in this monitoring task, thereby ensuring that subsequent video analysis is always carried out with the support of the algorithm that is most suitable for the current environmental conditions.

[0047] In one embodiment, after determining the target monitoring algorithm, the currently running algorithm is unloaded and the target monitoring algorithm is loaded and run.

[0048] Specifically, in order to achieve dynamic and seamless switching of algorithms and ensure that computing resources are always used optimally, after detecting a significant change in the scene and determining a new target monitoring algorithm, a termination command is first sent to the currently running old algorithm process, and it is waited for it to completely release the video memory, RAM and dedicated computing units it occupies. After the resources are released, the new target monitoring algorithm program is loaded from the algorithm library according to the command, the resources required for its operation are allocated to it, and it is started so that it can begin to receive and process subsequent video data. In this way, the optimal configuration of system resources is achieved while ensuring the monitoring effect.

[0049] In one embodiment, such as Figure 4 As shown, in step S30, the target behavior feature data is determined based on the target monitoring algorithm and video data, specifically including: S31: Use target detection algorithms to analyze video data and identify target objects in the video data.

[0050] Specifically, a defined target detection algorithm is used to process the video data collected in real time by the surveillance camera. First, the target detection function in the algorithm, such as the YOLO series model, is used to perform forward inference calculations on each frame of the image and output the location information of one or more target objects identified in the image. Target objects include pedestrians, vehicles, etc. The location information is usually the bounding box coordinates, as well as the category label and confidence score of each object.

[0051] S32: Associate the target object with the video data to generate the target object's motion trajectory data. Based on the motion trajectory data, determine the target object's target behavior characteristic data, which includes at least the movement speed and the duration of stay within the monitoring area.

[0052] Specifically, a multi-target tracker assigns a unique ID to each target object detected in the process. Objects with the same ID are associated across consecutive video frames, and their position coordinates in each frame are concatenated to form the object's motion trajectory data. Based on this trajectory data, the object's speed can be estimated in real time by calculating the displacement changes at adjacent time points. The total duration of the object's stay within the monitoring area can be calculated by recording the frame number or timestamp of the first and last occurrence of the ID. These data collectively constitute the target's behavioral characteristic data.

[0053] In one embodiment, such as Figure 5 As shown, in step S40, the anomaly monitoring strategy is determined based on the target behavior characteristic data, specifically including: S41: Construct a spatiotemporal behavior map based on target behavior feature data.

[0054] Specifically, each continuously tracked target object is treated as a dynamic node in the graph, while the monitoring scene is pre-divided into multiple virtual regions with different security attributes, which serve as static nodes. When a target object's trajectory enters or passes through a region, a weighted directed edge is established between the target object node and the corresponding region node. The edge weight can integrate information such as the object's movement speed, the duration of stay within the region, and the timestamp of entering the region, thereby integrating scattered behavioral feature data into a structured graph that reflects the spatiotemporal relationships of multiple targets.

[0055] S42: Input the spatiotemporal behavior map into the preset abnormal behavior classification model for analysis, and obtain the abnormal results from the abnormal behavior classification model.

[0056] Specifically, the abnormal behavior classification model can be a graph neural network model pre-trained with a spatiotemporal behavior map of a large number of normal behavior samples. This model can learn the normal patterns of human or vehicle behavior in different scenarios. When it receives the spatiotemporal behavior map constructed in real time, the model will analyze the structural features of the map and the interaction relationship between nodes, such as whether there are nodes staying in sensitive areas for too long, multiple nodes abnormally clustering, or movement speeds that do not match the attributes of the area. Finally, it will output one or more quantified abnormal intent labels and corresponding confidence scores.

[0057] S43: Based on the abnormal results, determine the anomaly monitoring strategy.

[0058] Specifically, the system has a pre-set policy mapping table that associates different abnormal results with specific countermeasures. For example, when the received abnormal result is that the target is loitering in a key area and the confidence level is higher than a preset alarm threshold, a high-priority monitoring policy is automatically matched and activated from the policy mapping table. This policy may include controlling the camera pan-tilt unit to automatically track the target object and perform optical zoom magnification, while pushing alarm information containing video clips before and after the event to the security personnel's mobile terminal.

[0059] In one embodiment, such as Figure 6 As shown, in step S42, which involves constructing a spatiotemporal behavior map based on the target behavior feature data, the specific steps include: S421: Divide the monitoring area into multiple area nodes and identify each target object in the video data as a target node.

[0060] Specifically, the entire monitoring scene can be pre-divided into several semantically meaningful sub-regions, such as entrance / exit areas, sensitive equipment areas, and public passage areas, by drawing polygons on a two-dimensional planar map of the monitoring screen through the management backend. Each divided polygon area corresponds to a static area node in the graph structure. For each active target in the video stream that is successfully tracked and assigned a unique identity ID, a dynamic target node is created for it in the graph structure. This node will persist until the target leaves the monitoring screen.

[0061] S422: Spatial mapping of the target object's motion trajectory data with multiple region nodes to generate a region node sequence and trajectory data.

[0062] Specifically, for each target node, the coordinates of its motion trajectory data at each time point are geometrically included with the polygonal range of all preset region nodes. When a coordinate point falls inside the polygon represented by a region node, the identifier of that region node is recorded. As the target object moves, its consecutive coordinate points are mapped to a sequence of region node identifiers ordered by time. This sequence describes the path the target object moves between various predefined regions.

[0063] S423: Based on trajectory data, construct association edges between target nodes and regional nodes, as well as between different regional nodes, to obtain a spatiotemporal behavior graph.

[0064] Specifically, when a target node stays within a certain region node for a duration exceeding a preset threshold, a weighted dwell edge is established between the target node and the region node, with its weight defined by the dwell duration. Simultaneously, based on the generated region node sequence, when a transition occurs from one region node to another adjacent region node, a weighted transfer edge is established between these two region nodes, with its weight defined by the frequency of such transfers or the average transfer time. By constructing these two types of edges, a complete spatiotemporal behavioral graph capable of simultaneously describing individual behavior and the flow of behavior within the scene is generated.

[0065] In one embodiment, such as Figure 7 As shown, in step S43, based on the abnormal results, the abnormality monitoring strategy is determined, which specifically includes: S431: If the confidence level in the abnormal result is less than the first threshold, then the abnormal monitoring strategy is determined to be the tracking and recording strategy.

[0066] Specifically, when the confidence level of identified abnormal behavior is low and does not reach the level requiring immediate human intervention, a silent background processing mechanism is activated. This mechanism labels the target and its related events with a low-risk tag and stores its complete motion trajectory data, associated spatiotemporal behavioral map snapshots, and information such as the specific abnormal intent and low confidence score output by the abnormal behavior classification model in the event log database. This information is used for subsequent data auditing, behavioral pattern analysis, or model iteration and optimization, thereby achieving the archiving and recording of potential risks without interfering with the normal work of security personnel. For example, two thresholds can be set: a first threshold of 0.6 and a second threshold of 0.9. If the confidence level of the abnormal result output by the model is less than 0.6, it indicates that the behavior is suspicious but the threat level is low. In this case, only the tracking and recording strategy is activated, the video clip of the target is tagged and archived for future reference.

[0067] S432: If the confidence level in the abnormal result is greater than the first threshold and less than the second threshold, then the abnormal monitoring strategy is determined to be the push alarm strategy.

[0068] Specifically, when the confidence level of abnormal behavior reaches a moderate level, indicating a clear potential risk, an automated alarm notification process will be triggered. This process will immediately capture real-time video footage containing the target object, and attach the object's identity ID, current location, detected abnormal behavior type, and corresponding confidence score. This information will be integrated into a structured alarm message and pushed via the network to the large-screen display system of the security monitoring center or the mobile devices of security personnel, reminding on-duty personnel to remotely review the video via sound or pop-up window. For example, if the confidence level is greater than 0.6 but less than 0.9, it indicates that the behavior is highly likely to constitute an anomaly. At this time, the push alarm strategy will be activated, pushing the alarm information along with the real-time video to the screen of the on-duty personnel in the monitoring center for manual review.

[0069] S433: If the confidence level in the abnormal result is greater than the second threshold, then the abnormal monitoring strategy is determined to be an on-site intervention strategy.

[0070] Specifically, when the confidence level of abnormal behavior exceeds the highest warning threshold, indicating a high probability of a security incident occurring or about to occur, the highest level of emergency response plan will be activated. At this time, not only will the highest level of audible and visual alarms be issued, but other security subsystems will also be automatically activated. For example, the access control system closest to the target's location will be automatically locked to restrict its movement, or the broadcast system will be invoked to issue a voice warning in a designated area. Simultaneously, a work order containing the optimal intervention path for the target will be generated and directly assigned to the security personnel closest to the incident site, continuously pushing the target's real-time location and video footage to their handheld terminals to support rapid and accurate on-site handling. If the confidence level is higher than 0.9, it means that high-risk abnormal behavior has been detected, and the system will automatically trigger the highest level of on-site intervention strategies, such as directly activating on-site audible and visual alarms, automatically locking doors, or notifying patrol personnel to handle the situation, thus forming a complete and intelligent closed-loop response system.

[0071] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data such as multimodal environment parameters, algorithm strategy libraries, multilayer perception models, and anomaly detection strategies. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a scene-adaptive anomaly detection method.

[0072] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Acquire multimodal environmental parameters within the monitoring area, and generate a feature vector for the current scene based on the multimodal environmental parameters; Based on the current scene feature vector, the target detection algorithm is determined from the preset algorithm strategy library; Acquire video data collected by camera equipment, and determine target behavior characteristic data based on target monitoring algorithms and video data; Based on the target behavior characteristic data, determine the anomaly monitoring strategy.

[0073] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Acquire multimodal environmental parameters within the monitoring area, and generate a feature vector for the current scene based on the multimodal environmental parameters; Based on the current scene feature vector, the target detection algorithm is determined from the preset algorithm strategy library; Acquire video data collected by camera equipment, and determine target behavior characteristic data based on target monitoring algorithms and video data; Based on the target behavior characteristic data, determine the anomaly monitoring strategy.

[0074] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0075] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0076] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A scene-adaptive anomaly detection method, characterized in that, The anomaly monitoring method includes: Acquire multimodal environmental parameters within the monitoring area, and generate a current scene feature vector based on the multimodal environmental parameters; Based on the current scene feature vector, a target detection algorithm is determined from a preset algorithm strategy library; Acquire video data collected by the camera device, and determine target behavior feature data based on the target monitoring algorithm and the video data; Based on the target behavior characteristic data, an anomaly monitoring strategy is determined.

2. The anomaly monitoring method according to claim 1, characterized in that, The step of generating the current scene feature vector based on the multimodal environment parameters specifically includes: The multimodal environmental parameters within the monitoring area are collected using optical sensors, infrared sensors, and a microphone array. These multimodal environmental parameters include at least light intensity data, environmental thermal distribution data, and background noise data. Based on the multimodal environment parameters, the current scene feature vector is determined through a preset multilayer perception model.

3. The anomaly monitoring method according to claim 2, characterized in that, The step of determining the target detection algorithm from a preset algorithm strategy library based on the current scene feature vector specifically includes: The current scene feature vector is matched with the algorithm strategy library to obtain the matching degree result of the monitoring algorithm; When the highest matching score in the matching results is greater than the preset matching threshold, the preset monitoring algorithm that best matches the algorithm strategy library is determined as the target monitoring algorithm.

4. The anomaly monitoring method according to claim 3, characterized in that, After determining the target monitoring algorithm, uninstall the currently running algorithm and load and run the target monitoring algorithm.

5. The anomaly monitoring method according to claim 1, characterized in that, The step of determining target behavior feature data based on the target monitoring algorithm and the video data specifically includes: The target detection algorithm is used to analyze the video data to determine the target objects in the video data; The target object is associated with the video data to generate motion trajectory data of the target object. Based on the motion trajectory data, the target behavior feature data of the target object is determined. The target behavior feature data includes at least the movement speed and the duration of stay in the monitoring area.

6. The anomaly monitoring method according to claim 5, characterized in that, The step of determining the anomaly monitoring strategy based on the target behavior feature data specifically includes: Based on the target behavior feature data, a spatiotemporal behavior map is constructed; The spatiotemporal behavior map is input into a preset abnormal behavior classification model for analysis, and abnormal results are obtained from the abnormal behavior classification model. Based on the abnormal results, the abnormality monitoring strategy is determined.

7. The anomaly monitoring method according to claim 6, characterized in that, The construction of a spatiotemporal behavior map based on the target behavior feature data specifically includes: The monitoring area is divided into multiple area nodes, and each target object in the video data is identified as a target node; The motion trajectory data of the target object is spatially mapped with multiple region nodes to generate a region node sequence and trajectory data. Based on the trajectory data, association edges are constructed between the target node and the regional node, as well as between different regional nodes, to obtain the spatiotemporal behavior graph.

8. The anomaly monitoring method according to claim 6, characterized in that, The step of determining the anomaly monitoring strategy based on the anomaly results specifically includes: If the confidence level in the abnormal result is less than the first threshold, then the abnormal monitoring strategy is determined to be a tracking and recording strategy. If the confidence level in the abnormal result is greater than the first threshold and less than the second threshold, then the abnormal monitoring strategy is determined to be a push alarm strategy. If the confidence level of the abnormal result is greater than the second threshold, then the abnormal monitoring strategy is determined to be an on-site intervention strategy.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the scene-adaptive anomaly detection method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the scene-adaptive anomaly detection method as described in any one of claims 1 to 8.

Images