Park safety monitoring method and system based on dynamic visual identification

Abstract

The invention relates to the technical field of monitoring analysis, in particular to a park safety monitoring method and system based on dynamic visual identification. The method comprises the following steps: acquiring a plurality of park monitoring videos based on park monitoring equipment; performing three-dimensional illumination migration rendering on the plurality of park monitoring videos, and performing low-frame smooth completion so as to generate a time sequence monitoring image frame sequence; carrying out spatial position positioning calculation on the park monitoring equipment, carrying out topological association network reconstruction, and constructing an omnibearing monitoring perception topological map; performing dynamic object visual detection on the time sequence monitoring image frame sequence, and performing frame-by-frame optical flow tracking analysis to generate a dynamic target optical flow trajectory; and carrying out multi-time-point sampling on the dynamic target optical flow trajectory, and then carrying out mutation point identification so as to generate an optical flow trajectory mutation point. According to the invention, efficient park safety monitoring analysis and high-precision safety early warning are realized.

Description

Technical Field

[0001] The present invention relates to the field of monitoring and analysis technology, and in particular to a park security monitoring method and system based on dynamic visual recognition. Background Art

[0002] With the acceleration of urbanization, park security issues have become increasingly prominent. Traditional park security monitoring methods usually rely on fixed cameras and manual inspections, which have problems such as limited monitoring range, slow response speed, and high cost. It is difficult to meet the increasingly complex park security management needs. In recent years, park security monitoring methods based on the Internet of Things and artificial intelligence technologies have attracted widespread attention. Using smart sensors, video analysis and other means, it can achieve all-round perception and intelligent analysis of key factors such as the internal environment of the park and personnel activities, providing strong technical support for park security management. However, how to effectively integrate and utilize these emerging technologies to establish a park security monitoring system covering the entire life cycle has become a key issue that needs to be solved urgently. Existing monitoring systems often have problems such as limited perception capabilities, low warning accuracy, and inaccurate monitoring analysis, making it difficult to adapt to the rapidly developing park security management needs. To this end, it is urgent to develop an innovative park security monitoring method based on dynamic perception and intelligent analysis, which can fully perceive changes in the internal environment of the park, quickly identify and warn of various safety hazards, and provide accurate and efficient decision-making support for park managers. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a campus security monitoring method and system based on dynamic visual recognition to solve at least one of the above technical problems.

[0004] To achieve the above object, the present invention provides a park security monitoring method based on dynamic visual recognition, comprising the following steps: Step S1: obtaining multiple park monitoring videos based on park monitoring equipment; performing three-dimensional lighting migration rendering on the multiple park monitoring videos, and performing low-frame smoothing completion, thereby generating a time-series monitoring image frame sequence; Step S2: Calculate the spatial location of the park monitoring equipment, reconstruct the topological association network, and build a comprehensive monitoring perception topology map; Step S3: Performing visual detection of dynamic objects on the time-series monitoring image frame sequence, and performing frame-by-frame optical flow tracking analysis to generate an optical flow trajectory of the dynamic target; Step S4: sampling the dynamic target optical flow trajectory at multiple time points, and then identifying the mutation points, thereby generating the optical flow trajectory mutation points; Step S5: Perform scene static building recognition on the time-series monitoring image frame sequence to generate park static scene structure data; perform abnormal safety risk analysis on the optical flow trajectory mutation points based on the park static scene structure data to obtain dynamic trajectory abnormal risk data; Step S6: Based on the dynamic trajectory abnormal risk data, abnormal targets are located on the omni-directional monitoring perception topology map, and abnormal safety prevention and control decisions are made to generate a campus safety decision engine.

[0005] The present invention improves the quality and continuity of monitoring videos through three-dimensional lighting migration rendering and low-frame smooth completion, providing more accurate data for subsequent analysis. The generated time-series monitoring image frame sequence helps to capture dynamic changes in the park monitoring video and lays the foundation for subsequent target analysis. The construction of an all-round monitoring perception topology map helps to understand the location layout and monitoring range of monitoring equipment, improves the overall efficiency of the monitoring system, and topological association network reconstruction helps to establish the connection relationship between devices, providing a basis for subsequent anomaly detection and positioning. Through dynamic object visual detection and optical flow tracking analysis, dynamic target information in monitoring videos is captured in real time, which helps to identify abnormal situations. The generated dynamic target optical flow trajectory provides important information for subsequent anomaly detection and risk analysis. Data support, by identifying the mutation points of optical flow trajectories, anomalies in the monitoring scene can be quickly captured, providing key clues for security risk analysis. The identification of mutation points helps to discover existing safety hazards in advance and enhance the effectiveness of park security monitoring. Through static building recognition and security risk analysis, a more comprehensive understanding of the environmental characteristics in the park can be achieved, which assists in the identification and analysis of abnormal situations. The generated dynamic trajectory abnormal risk data helps to assess the safety status of the park and provide data support for subsequent decision-making. Based on the analysis of abnormal risk data and the all-round monitoring perception topology map, abnormal targets can be located in time and corresponding security prevention and control decisions can be taken. The constructed park security decision-making engine helps to improve the response speed and decision-making efficiency of park security monitoring and ensure the safety of park residents.

[0006] Preferably, step S1 comprises the following steps: Step S11: Acquire multiple park monitoring videos based on park monitoring equipment; Step S12: performing video background texture feature analysis on multiple park surveillance videos to extract background texture features; Step S13: performing background noise analysis on the background texture features to obtain video background noise data; Step S14: performing multi-frequency filtering optimization on multiple park surveillance videos based on video background noise data to generate filtered noise-reduced surveillance videos; Step S15: performing three-dimensional illumination migration rendering on the filtered and denoised surveillance video to construct an illumination rendering surveillance video; Step S16: Perform low-frame smoothing on the illumination rendering monitoring video to generate a time-series monitoring image frame sequence.

[0007] The present invention obtains rich monitoring data by acquiring multiple park monitoring videos, and provides multi-angle information to maintain park safety. By analyzing the background texture features of the video, the background information in the monitoring video is identified, which helps to distinguish between the target and background elements. Background noise analysis helps to understand the noise situation in the monitoring video and provides a basis for subsequent noise reduction processing. Multi-frequency filtering optimization effectively reduces the noise in the video, improves the clarity and quality of the monitoring video, and helps to accurately monitor the target. Light migration rendering enhances the visual effect of the monitoring video, improves the viewing experience, and helps to present the monitoring scene more clearly. Low-frame smooth completion improves the fluency of the monitoring video, making the monitoring picture more coherent, which helps to reduce information loss and improve the monitoring effect.

[0008] Preferably, the specific steps of step S15 are: Extract multiple perspective images based on filtered denoising surveillance video; Identify brightness feature points of multiple viewing angle images and mark multiple viewing angle brightness feature points; Calculate the brightness of multiple viewing angle brightness feature points at the current timestamp to generate the brightness value of the image feature points; Identify brightness changes of image feature point brightness values ​​to obtain multi-view feature point brightness change data; Performing feature point spatial matching on multiple viewing angle brightness feature points to obtain a brightness feature point spatial matching relationship; Based on the spatial matching relationship of brightness feature points, the illumination intensity distribution data of multi-view feature points is analyzed to obtain the image illumination intensity distribution data; Perform three-dimensional illumination migration simulation on image illumination intensity distribution data to construct a dynamic three-dimensional illumination field; The park surveillance video is reconstructed by three-dimensional illumination field rendering according to the dynamic three-dimensional illumination field to construct the illumination rendering surveillance video.

[0009] The present invention obtains a more comprehensive monitoring screen by extracting images from multiple perspectives, which helps to comprehensively analyze the monitoring scene. Identifying and marking brightness feature points helps to locate key areas in the monitoring image and provide important information about the target area. By calculating the brightness value, the brightness changes of the image feature points are quantified to provide specific data support for subsequent analysis. Identifying brightness changes helps capture dynamic information in the monitoring scene and provide more accurate monitoring data. The establishment of spatial matching relationships between feature points helps to clarify the connection between brightness feature points of images from different perspectives and provide a more comprehensive data analysis perspective. Through the analysis of light intensity distribution, the lighting conditions of different areas in the monitoring scene can be understood to provide basic data for subsequent lighting rendering. Three-dimensional lighting migration simulation helps to simulate real lighting conditions and improve the visual effect and realism of the monitoring video. Light field rendering reconstruction enhances the lighting effect of the monitoring video, improves the visual quality of the picture, and helps to present the monitoring scene more clearly.

[0010] Preferably, the specific steps of step S16 are: Define the sliding time window length; Performing time window division processing on the illumination rendering surveillance video according to the length of the sliding time window, and extracting surveillance videos of multiple time windows; Perform dynamic frame sliding analysis on each frame of the surveillance image based on the time interval of all image frames to identify images in low frame rate areas; Perform convolution learning prediction on low frame rate area images to extract video context convolution information; Based on the video context convolution information, low-frame smooth completion is performed on the low-frame rate area image to generate a smooth frame completion video stream; Based on the smooth frame completion video stream, frame-by-frame extraction is performed to obtain multiple monitoring image frames; Calculate the timestamp of each surveillance image frame; Performing equal-interval encoding based on the timestamp to obtain a number for each monitoring image frame; Based on the serial number of each monitoring image frame, multiple video image frames are serialized in a time sequence standard to generate a time sequence monitoring image frame sequence.

[0011] The present invention helps to effectively cut and process monitoring video data by setting an appropriate time window length so that subsequent analysis and processing steps can run more effectively. Dividing the monitoring video into multiple time windows helps to segment the video data and improve processing efficiency and accuracy. Through dynamic frame sliding analysis and low frame rate area image recognition, the low frame rate area in the monitoring video is accurately captured to provide key information for subsequent processing. Based on the video context convolution information predicted and extracted by convolution learning, the characteristics and structure of the monitoring video data are better understood, and accurate information support is provided for completing the low frame rate area. Through low-frame smooth completion and generation of smooth frame completion video streams, the quality and continuity of the monitoring video are improved, and the video viewing experience is improved. Frame-by-frame extraction and timestamp calculation help to accurately mark the time information for each monitoring image frame, providing accurate time clues for subsequent timing analysis. Through equal-spacing encoding and timing standard serialization, a unified numbering system is established for monitoring image frames, which helps to integrate and compare monitoring data at different time points. Generating a time-series monitoring image frame sequence helps to integrate the information in the monitoring video, providing an important data basis for subsequent timing analysis, anomaly detection and security decision-making.

[0012] Preferably, step S2 specifically comprises the following steps: Step S21: performing spatial location calculation on the park monitoring equipment to obtain coordinates of multiple equipment location points; Step S22: performing spatial distribution analysis on the coordinates of multiple device location points to generate monitoring device spatial distribution data; Step S23: performing device space distance calculation on the coordinates of multiple device location points to obtain distance parameters between devices; Step S24: Calculate the visual detection range of multiple park monitoring videos to obtain the device monitoring visual area; Step S25: performing cross-monitoring topological area analysis on the device monitoring visual area based on the distance parameter between devices, thereby generating inter-device monitoring topological association data; Step S26: Based on the monitoring topology association data between devices, the monitoring device spatial distribution data is reconstructed into a topology association network to construct an all-round monitoring perception topology map.

[0013] The present invention helps to establish an accurate map of the device location by accurately calculating the location coordinates of each monitoring device, thereby improving the reliability and accuracy of the monitoring system. Through spatial location positioning, the layout of the monitoring equipment can be better understood, providing basic data for subsequent analysis and decision-making, and generating monitoring equipment spatial distribution data to reveal the coverage and density of the monitoring equipment in the park, helping to evaluate the overall monitoring situation. Through spatial distribution analysis, monitoring blind spots and overlapping areas are discovered, providing a reference basis for optimizing the monitoring layout. Through distance calculation between devices, the distance relationship between monitoring devices is determined, providing data support for subsequent cross-monitoring analysis. The acquisition of distance parameters between devices helps to evaluate the coverage and overlapping of monitoring devices, thereby improving the efficiency of the monitoring system. Through visual detection range calculation, the monitoring visual area of ​​each monitoring device is determined to help understand the monitoring range of the monitoring equipment. The determination of the equipment monitoring visual area is helpful to identify monitoring blind spots and overlapping areas, and provide guidance for the improvement of the monitoring system. Through cross-monitoring topological area analysis of the equipment monitoring visual area, the monitoring coverage and complementarity between monitoring devices are revealed. The generated monitoring topology association data between devices helps to optimize the monitoring layout and improve the overall efficiency of the monitoring system. By reconstructing the spatial distribution data and topological association network of monitoring equipment, a full-range monitoring perception topology map is constructed to improve the overall intelligence of the monitoring system. The reconstruction of the topological association network helps to establish the connection relationship between devices and provide support for comprehensive analysis and decision-making of security monitoring. Preferably, step S3 specifically comprises the following steps: Step S31: Perform visual detection of dynamic objects on the time-series monitoring image frame sequence and mark dynamic targets; Step S32: performing semantic recognition of the object type of the dynamic target to generate a dynamic object type, wherein the dynamic object type includes a vehicle and a pedestrian; Step S33: performing image segmentation of vehicles and pedestrians, and extracting vehicle target image frames and pedestrian target image frames; Step S34: performing frame-by-frame optical flow tracking analysis on the time-series monitoring image frame sequence according to the vehicle target image frame and the pedestrian target image frame to obtain the displacement vector of the dynamic image frame; Step S35: performing continuous frame displacement vector aggregation on the displacement vector of the dynamic image frame to generate a dynamic target optical flow trajectory.

[0014] The present invention effectively marks dynamic targets in monitoring images through dynamic object visual detection, and provides real-time perception of activities in the park. Marking dynamic targets helps to identify potential security threats or abnormal situations, and provides a basis for subsequent analysis and response. Object type semantic recognition of dynamic targets helps to determine whether the target is a vehicle or a pedestrian, and provides a basis for subsequent behavior analysis and trajectory tracking. Generating dynamic object type information helps to distinguish and analyze the behavior characteristics of different types of targets in the park. Image frames of vehicle targets and pedestrian targets are extracted through image labeling and segmentation, which helps to accurately identify and track targets. The extracted target image frame provides target position and boundary information for subsequent optical flow tracking analysis. Frame-by-frame optical flow tracking analysis captures the motion changes of the target between consecutive frames, and generates displacement vectors of the dynamic image frame. The displacement vector provides the motion trajectory information of the target, which helps to understand the motion mode and path of the target. Continuous frame displacement vector aggregation can generate the optical flow trajectory of the dynamic target, showing the motion path and behavior of the target in the monitoring image. The generation of the optical flow trajectory helps to analyze and infer the motion behavior of the target, and provides more information for the assessment of the security status of the park.

[0015] Preferably, the specific steps of step S4 are: Step S41: sampling the dynamic target optical flow trajectory at multiple time points to obtain multiple optical flow trajectory sampling points; Step S42: extracting the timestamp of the optical flow trajectory sampling point; Step S43: calculating the interval displacement speed of multiple optical flow trajectory sampling points based on the timestamp, so as to obtain the displacement speed between the sampling points; Step S44: performing multi-point speed change calculation on the displacement speed between sampling points to generate the speed change rate of the optical flow trajectory; Step S45: identifying mutation points of multiple optical flow trajectory sampling points based on the optical flow trajectory speed change rate, thereby generating optical flow trajectory mutation points.

[0016] The present invention captures the evolution process of the optical flow trajectory more comprehensively through multi-time point sampling, and provides more information about the target motion trajectory. Multiple sampling points reveal the position and motion state of the target at different time points, and provide rich data support for subsequent speed calculation and change rate analysis. The timestamp provides the specific time information of the sampling point, which is helpful for time series analysis and comparison of the trajectory data. The timestamp is used to calculate the time interval between the sampling points, and then derive the displacement speed and speed change rate. By calculating the interval displacement speed, the displacement of the target between different time points can be understood, and the target's motion speed and direction changes can be revealed. The displacement speed provides the target's motion dynamic information, laying the foundation for the subsequent speed change rate calculation. The multi-point speed change rate calculation reveals the changing trend of the target's motion speed, and helps understand the acceleration or deceleration of the target. The speed change rate reflects the dynamic change of the target's motion state, and provides a basis for the subsequent mutation point identification. By identifying the mutation point, the position where the speed or direction mutation exists in the trajectory is determined, marking an important change point of the target's motion state. The identification of the mutation point helps to capture the target's sudden behavior or abnormal movement, and provides important warning and analysis basis for park safety monitoring. Preferably, the specific steps of step S5 are: Step S51: performing scene static building recognition on the time-series monitoring image frame sequence, and marking the park scene static building data; Step S52: Performing scene layout structure analysis on the park scene static building data to generate park scene static structure data; Step S53: performing geometric distance calculation on the mutation point of the optical flow trajectory based on the static scene structure data of the park, and identifying the distance between the mutation point and the scene structure; Step S54: performing personalized behavior pattern analysis according to the distance between the mutation point and the scene structure to generate dynamic trajectory behavior pattern features; Step S55: performing abnormal safety risk analysis on the dynamic trajectory behavior pattern characteristics, thereby obtaining dynamic trajectory abnormal risk data.

[0017] The present invention marks the static building data of the park scene in the monitoring image through static building recognition, and provides scene environment information for subsequent analysis. The marking of static building data is helpful to understand the layout structure and scene characteristics in the park, and provides background information for analyzing dynamic trajectory behavior. The scene layout structure analysis generates the static scene structure data of the park, and reveals the layout relationship between the buildings and the environment in the park. The static scene structure data provides the scene background and reference for the subsequent distance calculation and behavior pattern analysis. The spatial relationship between the dynamic trajectory mutation point and the scene structure is quantified through geometric distance calculation, and the distance of the mutation point relative to the building is identified. The distance calculation helps to determine whether the mutation point is associated with a specific building or scene feature, and provides a basis for the subsequent behavior pattern analysis. The personalized behavior pattern analysis generates personalized behavior pattern features of the dynamic trajectory, such as detour, stay, etc., based on the distance between the mutation point and the scene structure. The generation of behavior pattern features helps to understand the specific behavior mode of the target in the park environment, and provides data support for abnormal risk analysis. The abnormal safety risk analysis identifies potential abnormal behaviors and safety risks based on the behavior pattern features of the dynamic trajectory, and provides targeted safety warnings. The generation of abnormal risk data of the dynamic trajectory helps to timely discover safety problems existing in the park, and provides important reference for safety monitoring decisions.

[0018] Preferably, the specific steps of step S6 are: Step S61: based on the dynamic trajectory abnormal risk data, reversely trace the dynamic target optical flow trajectory and mark the abnormal dynamic target; Step S62: quickly locate the target on the omni-directional monitoring perception topology map according to the abnormal dynamic target, so as to obtain the position coordinates of the abnormal target; Step S63: Perform intelligent warning response based on the position coordinates of the abnormal target and generate an intelligent warning signal; Step S64: Make abnormal safety prevention and control decisions based on the intelligent early warning signal to generate a campus safety decision engine.

[0019] The present invention uses reverse tracing to mark abnormal dynamic targets based on dynamic trajectory abnormal risk data, quickly identify targets with potential safety hazards, and marking abnormal dynamic targets helps targeted monitoring and analysis, thereby improving the efficiency of identifying and handling abnormal events. Rapid target positioning is based on abnormal dynamic targets, and the position coordinates of abnormal targets are determined on an all-round monitoring perception topology map to achieve positioning accuracy. Target positioning provides accurate target position information for subsequent response and processing, which helps to take corresponding measures in a timely manner. Based on the position coordinates of the abnormal targets, the intelligent early warning response can generate specific intelligent early warning signals to remind monitoring personnel or systems to perform further processing. Intelligent early warning signals help to quickly respond to abnormal events, reduce safety risks and reduce potential losses. Based on the intelligent early warning signals, the park security decision-making engine makes abnormal safety prevention and control decisions, including dispatching police force, initiating emergency measures, etc. The security decision-making engine improves the resilience and decision-making efficiency of the park security monitoring system through intelligent analysis and response.

[0020] In this specification, a campus security monitoring system based on dynamic visual recognition is provided, which is used to execute the campus security monitoring method based on dynamic visual recognition as described above, including: The video enhancement module is used to obtain multiple park monitoring videos based on the park monitoring equipment; perform three-dimensional lighting migration rendering on multiple park monitoring videos, and perform low-frame smoothing to generate a time-series monitoring image frame sequence; The monitoring perception topology module is used to calculate the spatial location of the park monitoring equipment, reconstruct the topological association network, and build a comprehensive monitoring perception topology map; The optical flow tracking module is used to perform visual detection of dynamic objects on the time-series monitoring image frame sequence and perform frame-by-frame optical flow tracking analysis to generate the optical flow trajectory of the dynamic target; The optical flow trajectory analysis portal is used to sample the optical flow trajectory of dynamic targets at multiple time points, and then identify the mutation points to generate the optical flow trajectory mutation points; The abnormal safety risk module is used to identify static buildings in the scene of the time-series monitoring image frame sequence and generate static scene structure data of the park; based on the static scene structure data of the park, the abnormal safety risk analysis of the optical flow trajectory mutation points is performed to obtain the dynamic trajectory abnormal risk data; The security decision-making module is used to locate abnormal targets in the all-round monitoring perception topology map based on dynamic trajectory abnormal risk data, and make abnormal security prevention and control decisions to generate a campus security decision-making engine.

[0021] The present invention can improve the quality and continuity of monitoring videos through three-dimensional illumination migration rendering and low-frame smooth completion, making the monitoring images clearer and more coherent. The generated time-series monitoring image frame sequence is helpful for subsequent analysis and processing, and provides high-quality input data for other modules. Through spatial position positioning calculation and topological association network reconstruction, an all-round monitoring perception topological map is constructed to fully display the spatial position relationship of the park monitoring equipment. The establishment of the topological map helps to improve the monitoring system's grasp of the overall situation of the park and provide basic support for subsequent positioning and decision-making. Through dynamic object visual detection and optical flow tracking analysis, the optical flow trajectory of the dynamic target can be generated to help the monitoring system track and record the moving path of the dynamic target in real time. The generation of the optical flow trajectory helps to analyze and predict the behavior of the dynamic target and improve the recognition of abnormal events. It has the ability to identify abnormal behaviors and events. By sampling multiple time points and identifying mutation points, it generates mutation points of optical flow trajectories, which helps to identify abnormal behaviors and events. Analyzing the mutation points of optical flow trajectories can help the system quickly locate potential abnormal targets and improve the response speed to abnormal situations. By identifying static buildings in the scene and analyzing the mutation points of optical flow trajectories, it generates dynamic trajectory abnormal risk data to help the system identify security risks in the park. Analyzing abnormal security risks helps to provide early warning and take corresponding safety prevention and control measures to ensure the safety of residents in the park. Based on the dynamic trajectory abnormal risk data, it locates abnormal targets and makes safety prevention and control decisions on the all-round monitoring perception topology map, generates a park safety decision-making engine, and improves the efficiency and accuracy of park safety management. The safety decision-making engine can quickly respond to abnormal events and help the system take timely countermeasures to ensure the safety of the park. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of the steps of a park security monitoring method based on dynamic visual recognition according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0023] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0024] The example of this application provides a park security monitoring method and system based on dynamic visual recognition. The execution subject of the park security monitoring method and system based on dynamic visual recognition includes but is not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of this application, and the data processing platform includes but is not limited to: at least one of an audio image management system, an information management system, and a cloud data management system.

[0025] See also Figures 1 to 4 The present invention provides a park security monitoring method based on dynamic visual recognition, and the park security monitoring method based on dynamic visual recognition comprises the following steps: Step S1: obtaining multiple park monitoring videos based on park monitoring equipment; performing three-dimensional lighting migration rendering on the multiple park monitoring videos, and performing low-frame smoothing completion, thereby generating a time-series monitoring image frame sequence; Step S2: Calculate the spatial location of the park monitoring equipment, reconstruct the topological association network, and build a comprehensive monitoring perception topology map; Step S3: Performing visual detection of dynamic objects on the time-series monitoring image frame sequence, and performing frame-by-frame optical flow tracking analysis to generate an optical flow trajectory of the dynamic target; Step S4: sampling the dynamic target optical flow trajectory at multiple time points, and then identifying the mutation points, thereby generating the optical flow trajectory mutation points; Step S5: Perform scene static building recognition on the time-series monitoring image frame sequence to generate park static scene structure data; perform abnormal safety risk analysis on the optical flow trajectory mutation points based on the park static scene structure data to obtain dynamic trajectory abnormal risk data; Step S6: Based on the dynamic trajectory abnormal risk data, abnormal targets are located on the omni-directional monitoring perception topology map, and abnormal safety prevention and control decisions are made to generate a campus safety decision engine.

[0026] The present invention improves the quality and continuity of monitoring videos through three-dimensional lighting migration rendering and low-frame smooth completion, providing more accurate data for subsequent analysis. The generated time-series monitoring image frame sequence helps to capture dynamic changes in the park monitoring video and lays the foundation for subsequent target analysis. The construction of an all-round monitoring perception topology map helps to understand the location layout and monitoring range of monitoring equipment, improves the overall efficiency of the monitoring system, and topological association network reconstruction helps to establish the connection relationship between devices, providing a basis for subsequent anomaly detection and positioning. Through dynamic object visual detection and optical flow tracking analysis, dynamic target information in monitoring videos is captured in real time, which helps to identify abnormal situations. The generated dynamic target optical flow trajectory provides important information for subsequent anomaly detection and risk analysis. Data support, by identifying the mutation points of optical flow trajectories, anomalies in the monitoring scene can be quickly captured, providing key clues for security risk analysis. The identification of mutation points helps to discover existing safety hazards in advance and enhance the effectiveness of park security monitoring. Through static building recognition and security risk analysis, a more comprehensive understanding of the environmental characteristics in the park can be achieved, which assists in the identification and analysis of abnormal situations. The generated dynamic trajectory abnormal risk data helps to assess the safety status of the park and provide data support for subsequent decision-making. Based on the analysis of abnormal risk data and the all-round monitoring perception topology map, abnormal targets can be located in time and corresponding security prevention and control decisions can be taken. The constructed park security decision-making engine helps to improve the response speed and decision-making efficiency of park security monitoring and ensure the safety of park residents.

[0027] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of the steps of a park security monitoring method based on dynamic visual recognition of the present invention. In this example, the steps of the park security monitoring method based on dynamic visual recognition include: Step S1: obtaining multiple park monitoring videos based on park monitoring equipment; performing three-dimensional lighting migration rendering on the multiple park monitoring videos, and performing low-frame smoothing completion, thereby generating a time-series monitoring image frame sequence; In this embodiment, the location and function of all monitoring devices in the park are determined, and information such as device ID, video format, and resolution are obtained. Video streams are obtained from each monitoring device in real time to ensure the continuity and integrity of the video. The acquired multiple park monitoring videos are stored in a unified data management system for subsequent processing. The acquired video is preliminarily analyzed, including frame extraction and illumination feature point detection. Static objects and dynamic targets in the scene are identified, and the characteristics of illumination changes are recorded. Based on the scene information in the monitoring video, a three-dimensional scene model is constructed, taking into account the illumination conditions and scene layout. A suitable illumination migration algorithm (such as a local illumination model or a global illumination model) is selected to calculate the three-dimensional illumination field. The three-dimensional illumination field is applied to the monitoring video to generate a realistic illumination effect and perform dynamic rendering. The extracted frame sequence is analyzed to identify low frame rate areas and determine which areas need to be completed. A suitable low-frame completion algorithm (such as a generative adversarial network or interpolation method based on deep learning) is selected. The identified low-frame areas are smoothly completed to ensure the smoothness and visual consistency of the video.

[0028] Step S2: Calculate the spatial location of the park monitoring equipment, reconstruct the topological association network, and build a comprehensive monitoring perception topology map; In this embodiment, basic information of all park monitoring devices is collected, including device ID, installation location (longitude, latitude, altitude), installation angle, etc., and a suitable positioning method (such as GPS positioning, triangulation, vision-based positioning, etc.) is selected. The selected positioning method is used to calculate the spatial position of each monitoring device to obtain the three-dimensional coordinates of the device, and the location data of each device is recorded to ensure the integrity and accuracy of the data. According to the spatial position of the device, the relative position relationship between the devices (such as distance, field of view overlap, etc.) is analyzed to determine which devices can monitor each other and form an adjacency relationship. A suitable network model (such as a graph model) is selected to define nodes (monitoring devices) and edges (monitoring relationships between devices) in the network. According to the adjacency relationship, a topological network of monitoring devices is constructed, and the connection relationship and monitoring coverage area of ​​each device are recorded. The constructed topological network data is saved in a visual format (such as an adjacency matrix or an edge list) for subsequent display and analysis. The structure of the topological map is designed to ensure that the spatial distribution and mutual relationship of the monitoring devices can be displayed. A suitable visualization tool (such as GIS software or a graphics library) is selected to draw the topological map, and the spatial position and topological association information of the monitoring devices are imported into the visualization tool to generate a full-range monitoring perception topological map.

[0029] Step S3: Performing visual detection of dynamic objects on the time-series monitoring image frame sequence, and performing frame-by-frame optical flow tracking analysis to generate an optical flow trajectory of the dynamic target; In this embodiment, video data is loaded from a time-series monitoring image frame sequence to ensure the quality and integrity of each frame of the image. A suitable dynamic object detection algorithm (such as YOLO, SSD or Faster R-CNN) is selected. These algorithms can detect and classify dynamic targets in real time. Dynamic object detection is performed on each frame of the image, and dynamic targets (such as vehicles, pedestrians, etc.) are identified and marked. The position information of each detected dynamic target is recorded, including its bounding box coordinates, category and confidence. A suitable optical flow algorithm (such as Lucas-Kanade method or Horn-Schunck method) is selected to calculate the displacement of the target in consecutive frames. From the extracted dynamic target information, the target to be tracked is determined, and an optical flow tracker is initialized for each target. For each frame in the time-series monitoring image frame sequence, the optical flow algorithm is applied to calculate the displacement vector of each dynamic target, the current position of each target is updated, the position change of the target in each frame is recorded, and the position change of each dynamic target in the consecutive frames is recorded as an optical flow trajectory to form a motion path of the target.

[0030] Step S4: sampling the dynamic target optical flow trajectory at multiple time points, and then identifying the mutation points, thereby generating the optical flow trajectory mutation points; In this embodiment, the sampling time interval (such as every N frames or every fixed time) is determined to ensure uniform coverage of the optical flow trajectory. According to the set sampling strategy, multiple sampling points are extracted from the optical flow trajectory, and the position information and timestamp of each sampling point are recorded so that the sampling point can represent the overall characteristics of the trajectory and ensure that important dynamic changes are captured. The sampled optical flow trajectory points and their corresponding timestamps are saved as a data set to facilitate subsequent mutation point analysis, and the recognition standard of the mutation point is determined. For example, a threshold is selected to determine whether the displacement change rate or the speed change rate is significant, and the displacement, speed and its change rate are obtained from the extracted multi-time point sampling data. According to the set standard, the changes between the sampling points are analyzed, and the points whose change rate exceeds the threshold are identified. Each identified mutation point is recorded, including its position and time information.

[0031] Step S5: Perform scene static building recognition on the time-series monitoring image frame sequence to generate park static scene structure data; perform abnormal safety risk analysis on the optical flow trajectory mutation points based on the park static scene structure data to obtain dynamic trajectory abnormal risk data; In this embodiment, a suitable building recognition algorithm (such as YOLO, Faster R-CNN or SegNet) is selected for static building detection. Static building recognition is performed on each frame of the image to identify all buildings in the park scene. The location information of each building is recorded, including its bounding box, type, height and other features. The identified building information is integrated to generate the park static scene structure data, including the relative position, layout and architectural features between buildings. A suitable abnormal safety risk analysis model (such as a rule-based model, a machine learning model or a statistical analysis method) is selected. The mutation points of the optical flow trajectory are analyzed, and the relationship between the mutation points and the buildings is evaluated in combination with the static scene structure data of the park. Identify potential safety risks, for example, whether the mutation point is close to an important building, whether it is in a high-risk area, etc. The risk level and related building information of each mutation point are recorded to form a dynamic trajectory abnormal risk data set.

[0032] Step S6: Based on the dynamic trajectory abnormal risk data, abnormal targets are located on the omni-directional monitoring perception topology map, and abnormal safety prevention and control decisions are made to generate a campus safety decision engine.

[0033] In this embodiment, the previously generated all-round monitoring perception topology map is loaded to ensure that it contains the location information of all monitoring equipment and buildings, and a suitable positioning algorithm (such as a location-based weighted algorithm or a regional clustering algorithm) is selected to locate the abnormal target. According to the abnormal trajectory data and the topology map, the specific location of the abnormal target in the monitoring area is quickly located to ensure that the positioning result is accurate, and the coordinates and related information of each abnormal target are recorded. A suitable decision support system (such as a rule-based decision tree, optimization algorithm or machine learning model) is selected to generate prevention and control decisions. According to the positioning information and risk level of the abnormal target, the corresponding security risk is evaluated, the nature and severity of the potential threat are determined, and specific prevention and control measures are formulated for the identified abnormal targets, such as: sending an alarm to the relevant monitoring equipment, requiring real-time attention to the abnormal target, notifying the park security personnel to conduct on-site inspections, starting the automatic monitoring system, recording the subsequent dynamics of the target, integrating the formulated prevention and control strategies and execution plans, forming a park security decision engine, and ensuring the automation and intelligence of the decision-making process.

[0034] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Step S11: Acquire multiple park monitoring videos based on park monitoring equipment; Step S12: performing video background texture feature analysis on multiple park surveillance videos to extract background texture features; Step S13: performing background noise analysis on the background texture features to obtain video background noise data; Step S14: performing multi-frequency filtering optimization on multiple park surveillance videos based on video background noise data to generate filtered noise-reduced surveillance videos; Step S15: performing three-dimensional illumination migration rendering on the filtered and denoised surveillance video to construct an illumination rendering surveillance video; Step S16: Perform low-frame smoothing on the illumination rendering monitoring video to generate a time-series monitoring image frame sequence.

[0035] In this embodiment, the park monitoring equipment (such as a camera) is determined to ensure that it has video recording and real-time monitoring functions. Configure the equipment to obtain monitoring videos of multiple park areas and set appropriate recording duration and resolution. Store the collected video data in a central server or cloud storage to ensure data security and integrity. Use a background modeling algorithm (such as a Gaussian mixture model) to separate the foreground and background in the video. Use texture analysis methods (such as grayscale co-occurrence matrix, local binary pattern) to extract the texture features of the background. Calculate relevant feature indicators (such as contrast, brightness, directionality, etc.). Analyze the noise components in the video. Calculate the noise level of each frame of the image, and use statistical methods (such as mean, variance) to evaluate the background noise. Generate background noise data and identify the type of noise (such as random noise, periodic noise). According to the background noise analysis results, select a suitable multi-frequency filter (such as wavelet transform, Gaussian filter). Apply multi-frequency filtering technology to denoise each monitoring video. Adjust the filtering parameters to optimize the video quality and retain important details as much as possible. Select a suitable illumination model (such as Phong model or Blinn-Phong model) for illumination rendering. Construct a 3D scene based on filtered and denoised surveillance video and extract geometric information from the scene. Apply a 3D illumination migration algorithm to perform illumination rendering on surveillance video. Generate realistic illumination effects to improve the visual performance of the video. Extract key frames from the illumination-rendered surveillance video and analyze the changes between frames. Select a suitable low-frame smooth completion algorithm (such as optical flow method, interpolation method) for completion processing. Perform time-series completion on the video to generate a smooth frame sequence to reduce the jump between frames. Calculate the completed frame sequence and ensure that each frame transitions naturally. Generate the final time-series surveillance image frame sequence and save it as a standard video format or image sequence.

[0036] In this embodiment, the specific steps of step S15 are: Extract multiple perspective images based on filtered denoising surveillance video; Identify brightness feature points of multiple viewing angle images and mark multiple viewing angle brightness feature points; Calculate the brightness of multiple viewing angle brightness feature points at the current timestamp to generate the brightness value of the image feature points; Identify brightness changes of image feature point brightness values ​​to obtain multi-view feature point brightness change data; Performing feature point spatial matching on multiple viewing angle brightness feature points to obtain a brightness feature point spatial matching relationship; Based on the spatial matching relationship of brightness feature points, the illumination intensity distribution data of multi-view feature points is analyzed to obtain the image illumination intensity distribution data; Perform three-dimensional illumination migration simulation on image illumination intensity distribution data to construct a dynamic three-dimensional illumination field; The park surveillance video is reconstructed by three-dimensional illumination field rendering according to the dynamic three-dimensional illumination field to construct the illumination rendering surveillance video.

[0037] In this embodiment, according to the layout of the monitoring scene, multiple perspectives (such as different camera perspectives) are selected for extraction, key frames are extracted from each perspective, and multiple perspective images are generated, ensuring that the images of each perspective have the same timestamp, and the multiple perspective images are processed using a feature point detection algorithm (such as Harris corner detection, SIFT or SURF) to identify and extract brightness feature points, mark the identified brightness feature points in each perspective image for subsequent analysis, and store the marked feature point information as a data set, including the feature point coordinates and descriptors of each perspective, ensuring that the extracted multiple perspective images have the same timestamp for brightness calculation, and calculating the brightness value of each feature point, usually using the average value of the grayscale value or the RGB value, to generate a brightness value data set of the image feature points, record the brightness value of each feature point at different perspectives, compare the brightness values ​​of the same feature point at different timestamps or different perspectives, identify brightness changes, generate multi-perspective feature point brightness change data, record the brightness change of each feature point (such as the increase or decrease, the rate of change), and store the brightness change data as a report. To facilitate subsequent analysis and processing, select a suitable spatial matching algorithm (such as RANSAC, KNN or optical flow method) to match the brightness feature points of multiple perspectives, determine the spatial matching relationship of the brightness feature points between different perspectives by comparing the descriptors and positions of the feature points, save the matching relationship data for subsequent analysis, record the corresponding perspective information of each matching point, calculate the illumination intensity distribution of the multi-perspective feature points according to the spatial matching relationship of the brightness feature points, generate image illumination intensity distribution data, record the illumination intensity (such as average intensity, maximum intensity) at different positions, create a three-dimensional model, use the illumination migration algorithm (such as the illumination transfer equation) to simulate the three-dimensional illumination field based on the extracted feature points and illumination intensity distribution data, run the illumination migration simulation, generate a dynamic three-dimensional illumination field according to the illumination intensity distribution, save the dynamic three-dimensional illumination field data for subsequent rendering, load the three-dimensional illumination field data and the campus surveillance video, select a suitable rendering algorithm (such as ray tracing or rasterization) to reconstruct the surveillance video, use the dynamic three-dimensional illumination field to perform illumination rendering on the surveillance video, and generate an illumination rendering surveillance video.

[0038] In this embodiment, the specific steps of step S16 are: Define the sliding time window length; Performing time window division processing on the illumination rendering surveillance video according to the length of the sliding time window, and extracting surveillance videos of multiple time windows; Perform dynamic frame sliding analysis on each frame of the surveillance image based on the time interval of all image frames to identify images in low frame rate areas; Perform convolution learning prediction on low frame rate area images to extract video context convolution information; Based on the video context convolution information, low-frame smooth completion is performed on the low-frame rate area image to generate a smooth frame completion video stream; Based on the smooth frame completion video stream, frame-by-frame extraction is performed to obtain multiple monitoring image frames; Calculate the timestamp of each surveillance image frame; Performing equal-interval encoding based on the timestamp to obtain a number for each monitoring image frame; Based on the serial number of each monitoring image frame, multiple video image frames are serialized in a time sequence standard to generate a time sequence monitoring image frame sequence.

[0039] In this embodiment, according to the monitoring requirements and the video frame rate, a suitable sliding time window length (for example, 5 seconds or 10 seconds) is defined, and the selected window length is used as a parameter to load video data from the illumination rendering monitoring video. According to the defined sliding time window length, the video is divided into multiple time windows, and the video in each time window is extracted to ensure the data integrity of each window. The image frames in each time window are dynamically analyzed, the time interval of each frame is calculated, and a low frame rate threshold is set (for example, less than 50% of the normal frame rate). The low frame rate area image is identified and marked, and its position in the time window is recorded. A suitable convolutional neural network (CNN) model is selected for image prediction, and the low frame rate area image is used as input data for convolution learning. The video context convolution information is extracted through the model, and the patterns and features in the image are identified. The extracted context convolution information is stored as a data set for subsequent completion processing, and a suitable smooth completion is selected. An algorithm (such as one based on a generative adversarial network or an interpolation method) is used to smoothly complete the image in the low frame rate area based on the extracted video context convolution information, create a smooth frame completion video stream, ensure that the video transition is natural and smooth, extract the monitoring image frame frame by frame from the smooth frame completion video stream, save each extracted monitoring image frame as an independent image file for subsequent processing, extract the time information of each monitoring image frame from the smooth completion video stream, calculate the timestamp for each frame according to the frame rate of the video and the extraction order, ensure that the time information is accurate, associate the timestamp information with each monitoring image frame to form a timestamp data set, perform equidistant encoding based on the timestamp of each monitoring image frame, generate a unique number for each frame, associate the generated number information with the monitoring image frame to form a number data set, perform time standard serialization on multiple video image frames according to the number of each monitoring image frame, create a time sequence monitoring image frame sequence, and ensure that the frames in the sequence are arranged in chronological order.

[0040] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Step S21: performing spatial location calculation on the park monitoring equipment to obtain coordinates of multiple equipment location points; Step S22: performing spatial distribution analysis on the coordinates of multiple device location points to generate monitoring device spatial distribution data; Step S23: performing device space distance calculation on the coordinates of multiple device location points to obtain distance parameters between devices; Step S24: Calculate the visual detection range of multiple park monitoring videos to obtain the device monitoring visual area; Step S25: performing cross-monitoring topological area analysis on the device monitoring visual area based on the distance parameter between devices, thereby generating inter-device monitoring topological association data; Step S26: Based on the monitoring topology association data between devices, the monitoring device spatial distribution data is reconstructed into a topology association network to construct an all-round monitoring perception topology map.

[0041] In this embodiment, relevant information of the park monitoring equipment is collected, including the equipment ID, installation location and installation angle, and a suitable positioning algorithm (such as GPS positioning, triangulation positioning method, etc.) is selected to perform spatial position calculation. The spatial position of each device is calculated using the positioning algorithm to obtain the coordinate points (such as longitude, latitude, and altitude) of multiple devices. A spatial distribution analysis method is selected, such as a Voronoi diagram or a K-Means cluster analysis, to perform spatial distribution analysis on the equipment location points, to identify the distribution characteristics of the equipment in space, to generate spatial distribution data of the monitoring equipment, to record the distribution density and regional characteristics of the equipment, to select a suitable distance calculation method (such as Euclidean distance or Manhattan distance) to calculate the distance between the equipment, to calculate the spatial distance between multiple equipment location points pair by pair, to generate distance parameter data, to record the distance between each pair of equipment, to form a distance parameter table between equipment, and to calculate the distance between the equipment according to the type and installation of the monitoring equipment. Parameters, select a suitable visual range model (such as circle, sector), calculate the visual detection range of each monitoring device, consider the viewing angle and distance of the device, generate visual area data for device monitoring, record the monitoring coverage and boundaries of each device, store the visual area data as a report, and facilitate association with other data. Based on the distance parameters and visual detection range between devices, identify the cross-monitoring areas between devices, use topological analysis methods (such as adjacency matrix or graph theory methods) to generate monitoring topology association data between devices, record the cross-monitoring relationship between devices, generate monitoring topology association data sets, describe the mutual coverage between devices, select a suitable network reconstruction model (such as a graph network or spatial network model) to reconstruct the topological association network, combine spatial distribution data with topological association data, build a complete monitoring network, generate a full range of monitoring perception topology maps, and display the spatial distribution and mutual relationship of devices.

[0042] In this embodiment, refer to Figure 4, is a flowchart of detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Step S31: Perform visual detection of dynamic objects on the time-series monitoring image frame sequence and mark dynamic targets; Step S32: performing semantic recognition of the object type of the dynamic target to generate a dynamic object type, wherein the dynamic object type includes a vehicle and a pedestrian; Step S33: performing image segmentation of vehicles and pedestrians, and extracting vehicle target image frames and pedestrian target image frames; Step S34: performing frame-by-frame optical flow tracking analysis on the time-series monitoring image frame sequence according to the vehicle target image frame and the pedestrian target image frame to obtain the displacement vector of the dynamic image frame; Step S35: performing continuous frame displacement vector aggregation on the displacement vector of the dynamic image frame to generate a dynamic target optical flow trajectory.

[0043] In this embodiment, a suitable dynamic object detection algorithm (such as YOLO, SSD or Faster R-CNN) is selected for real-time detection, dynamic object detection is performed on each frame of the image, dynamic targets (such as vehicles and pedestrians) are identified and marked, the location information of each detected dynamic target, including coordinates and bounding boxes, is recorded, a suitable deep learning model (such as convolutional neural network) is selected for object type recognition, the detected dynamic targets are used as input data, object type recognition is performed, the detected dynamic targets (such as vehicles and pedestrians) are classified, the corresponding dynamic object types are generated, the type information of each dynamic target is recorded, the accuracy of the data is ensured, and a suitable image segmentation algorithm (such as Mask Use R-CNN or U-Net) to perform target segmentation, perform image labeling and segmentation on the identified vehicles and pedestrians, extract target image frames, generate accurate image frames for each target, ensure that the frames contain complete target information, save the position information and segmentation results of the vehicle target image frames and the pedestrian target image frames, select a suitable optical flow algorithm (such as Lucas-Kanade or Horn-Schunck) for frame-by-frame analysis, perform optical flow tracking based on the target image frames of vehicles and pedestrians for each frame in the time-series monitoring image frame sequence, calculate the displacement vector of each target in the continuous frames, record the movement of the target, select a suitable aggregation method (such as average aggregation or weighted aggregation) according to the characteristics of the target movement, aggregate the displacement vectors of multiple continuous frames, generate optical flow trajectories of dynamic targets, calculate the motion trajectory of each target, and record its movement path in the monitoring area. In this embodiment, step S4 includes the following steps: Step S41: sampling the dynamic target optical flow trajectory at multiple time points to obtain multiple optical flow trajectory sampling points; Step S42: extracting the timestamp of the optical flow trajectory sampling point; Step S43: calculating the interval displacement speed of multiple optical flow trajectory sampling points based on the timestamp, so as to obtain the displacement speed between the sampling points; Step S44: performing multi-point speed change calculation on the displacement speed between sampling points to generate the speed change rate of the optical flow trajectory; Step S45: identifying mutation points of multiple optical flow trajectory sampling points based on the optical flow trajectory speed change rate, thereby generating optical flow trajectory mutation points.

[0044] In this embodiment, a sampling interval (such as every 1 second or every N frames) is determined for sampling to ensure that the sampling points cover the entire trajectory. According to the defined sampling strategy, multiple optical flow trajectory sampling points are extracted from the optical flow trajectory, the position information of each sampling point is recorded, and the sampling point information (including coordinates and timestamps) is stored as a data set for subsequent analysis. The timestamp of each sampling point is extracted, and the timestamp corresponding to each sampling point is recorded to ensure that the time information is complete and accurate. The timestamp information is associated with the corresponding sampling point to form a timestamp data set. For each pair of adjacent optical flow trajectory sampling points, the displacement between them (that is, the distance between the two points) is calculated, and the time interval is calculated using the extracted timestamp, and then the interval displacement speed of each pair of sampling points is calculated. Each pair of adjacent speed values ​​in the speed data set is compared to calculate the change rate between them, and the speed change of each sampling point is recorded to generate optical flow trajectory speed change rate data, and the recognition standard of the mutation point is determined (such as the change rate exceeds a certain threshold). According to the calculated speed change rate, the mutation point, that is, the point with a significant change rate, is identified, and the identified mutation point and its corresponding sampling point information are recorded.

[0045] In this embodiment, the specific steps of step S5 are: Step S51: performing scene static building recognition on the time-series monitoring image frame sequence, and marking the park scene static building data; Step S52: Performing scene layout structure analysis on the park scene static building data to generate park scene static structure data; Step S53: performing geometric distance calculation on the mutation point of the optical flow trajectory based on the static scene structure data of the park, and identifying the distance between the mutation point and the scene structure; Step S54: performing personalized behavior pattern analysis according to the distance between the mutation point and the scene structure to generate dynamic trajectory behavior pattern features; Step S55: performing abnormal safety risk analysis on the dynamic trajectory behavior pattern characteristics, thereby obtaining dynamic trajectory abnormal risk data.

[0046] In this embodiment, video data is loaded from a time-series monitoring image frame sequence, a suitable building detection algorithm (such as YOLO or Faster R-CNN) is selected for static building recognition, static building detection is performed on each frame of the image, the buildings in the campus scene are identified and marked, the location information of each building, including coordinates and bounding boxes, is recorded, a suitable analysis method (such as Delaunay triangulation or Voronoi diagram) is selected for scene layout structure analysis, the layout structure data of the campus scene is generated by analyzing the location and shape of the buildings, the relative position relationship and spatial distribution characteristics between the buildings are recorded, a suitable geometric distance calculation method (such as Euclidean distance or Manhattan distance) is selected for distance calculation between the mutation point and the scene structure, the geometric distance between each mutation point and the scene structure is calculated according to the identified mutation point of the optical flow trajectory, and the distance between each mutation point and its nearest building is recorded Or the distance information between scene structures, select a suitable behavior pattern analysis model (such as cluster analysis or hidden Markov model) for personalized behavior pattern recognition, extract relevant features for analysis based on the calculated distance data and optical flow trajectory data, analyze the behavior pattern of the mutation point, identify the individual's motion characteristics and patterns in the scene, generate dynamic trajectory behavior pattern features, record the behavior feature information of each mutation point, select a suitable anomaly detection algorithm (such as isolation forest or autoencoder) for security risk analysis, use the generated dynamic trajectory behavior pattern features as input data, perform anomaly detection on the dynamic trajectory behavior pattern features, identify potential security risks, and generate dynamic trajectory abnormal risk data, including the type and risk level of abnormal behavior.

[0047] In this embodiment, the specific steps of step S6 are: Step S61: based on the dynamic trajectory abnormal risk data, reversely trace the dynamic target optical flow trajectory and mark the abnormal dynamic target; Step S62: quickly locate the target on the omni-directional monitoring perception topology map according to the abnormal dynamic target, so as to obtain the position coordinates of the abnormal target; Step S63: Perform intelligent warning response based on the position coordinates of the abnormal target and generate an intelligent warning signal; Step S64: Make abnormal safety prevention and control decisions based on the intelligent early warning signal to generate a campus safety decision engine.

[0048] In this embodiment, potential abnormal dynamic targets are identified from the acquired dynamic trajectory abnormal risk data, and a suitable reverse tracing algorithm (such as a backtracking algorithm based on time series analysis) is selected. For the identified abnormal dynamic targets, their optical flow trajectories are analyzed, their movement history in the monitoring area is traced, key points and events of the abnormal dynamic targets on the trajectory are marked, their behavioral characteristics are recorded, and a full range of monitoring perception topological maps are loaded. A suitable fast target positioning algorithm (such as a particle filter or a Kalman filter) is selected. According to the marked abnormal dynamic target information, the topological map is used for rapid positioning, and the specific position coordinates of the abnormal target in the monitoring area are calculated to ensure the accuracy of positioning. A suitable intelligent early warning signal generation model (such as a threshold-based alarm system) is selected. Based on the position coordinates of the abnormal target and the identified behavioral characteristics, an intelligent early warning signal is generated, the type of the early warning signal (such as an emergency alarm, a warning signal, etc.) is determined, and the corresponding response level is set. A suitable decision support system (such as a rule-based decision tree or a machine learning model) is selected. Based on the intelligent early warning signal, the potential risks and impacts of the abnormal targets are analyzed, and corresponding park safety prevention and control decisions are generated, including emergency measures, alarm notifications, or on-site response strategies. The park safety decisions are recorded as the output of the decision engine for subsequent execution and evaluation.

[0049] In this embodiment, a campus security monitoring system based on dynamic visual recognition is provided, which is used to execute the campus security monitoring method based on dynamic visual recognition as described above, including: The video enhancement module is used to obtain multiple park monitoring videos based on the park monitoring equipment; perform three-dimensional lighting migration rendering on multiple park monitoring videos, and perform low-frame smoothing to generate a time-series monitoring image frame sequence; The monitoring perception topology module is used to calculate the spatial location of the park monitoring equipment, reconstruct the topological association network, and build a comprehensive monitoring perception topology map; The optical flow tracking module is used to perform visual detection of dynamic objects on the time-series monitoring image frame sequence and perform frame-by-frame optical flow tracking analysis to generate the optical flow trajectory of the dynamic target; The optical flow trajectory analysis portal is used to sample the optical flow trajectory of dynamic targets at multiple time points, and then identify the mutation points to generate the optical flow trajectory mutation points; The abnormal safety risk module is used to identify static buildings in the scene of the time-series monitoring image frame sequence and generate static scene structure data of the park; based on the static scene structure data of the park, the abnormal safety risk analysis of the optical flow trajectory mutation points is performed to obtain the dynamic trajectory abnormal risk data; The security decision-making module is used to locate abnormal targets in the all-round monitoring perception topology map based on dynamic trajectory abnormal risk data, and make abnormal security prevention and control decisions to generate a campus security decision-making engine.

[0050] The present invention can improve the quality and continuity of monitoring videos through three-dimensional illumination migration rendering and low-frame smooth completion, making the monitoring images clearer and more coherent. The generated time-series monitoring image frame sequence is helpful for subsequent analysis and processing, and provides high-quality input data for other modules. Through spatial position positioning calculation and topological association network reconstruction, an all-round monitoring perception topological map is constructed to fully display the spatial position relationship of the park monitoring equipment. The establishment of the topological map helps to improve the monitoring system's grasp of the overall situation of the park and provide basic support for subsequent positioning and decision-making. Through dynamic object visual detection and optical flow tracking analysis, the optical flow trajectory of the dynamic target can be generated to help the monitoring system track and record the moving path of the dynamic target in real time. The generation of the optical flow trajectory helps to analyze and predict the behavior of the dynamic target and improve the recognition of abnormal events. It has the ability to identify abnormal behaviors and events. By sampling multiple time points and identifying mutation points, it generates mutation points of optical flow trajectories, which helps to identify abnormal behaviors and events. Analyzing the mutation points of optical flow trajectories can help the system quickly locate potential abnormal targets and improve the response speed to abnormal situations. By identifying static buildings in the scene and analyzing the mutation points of optical flow trajectories, it generates dynamic trajectory abnormal risk data to help the system identify security risks in the park. Analyzing abnormal security risks helps to provide early warning and take corresponding safety prevention and control measures to ensure the safety of residents in the park. Based on the dynamic trajectory abnormal risk data, it locates abnormal targets and makes safety prevention and control decisions on the all-round monitoring perception topology map, generates a park safety decision-making engine, and improves the efficiency and accuracy of park safety management. The safety decision-making engine can quickly respond to abnormal events and help the system take timely countermeasures to ensure the safety of the park.

[0051] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0052] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A park security monitoring method based on dynamic visual recognition, characterized in that: The following steps are involved: Step S1: Acquire multiple park monitoring videos based on park monitoring equipment; Perform 3D lighting migration rendering on multiple campus surveillance videos and perform low-frame smoothing to generate a time-series surveillance image frame sequence; Step S2: Calculate the spatial location of the park monitoring equipment, reconstruct the topological association network, and build a comprehensive monitoring perception topology map; Step S3: Performing visual detection of dynamic objects on the time-series monitoring image frame sequence, and performing frame-by-frame optical flow tracking analysis to generate an optical flow trajectory of the dynamic target; Step S4: sampling the dynamic target optical flow trajectory at multiple time points, and then identifying the mutation points, thereby generating the optical flow trajectory mutation points; Step S5: Perform scene static building recognition on the time-series monitoring image frame sequence to generate park static scene structure data; Based on the static scene structure data of the park, abnormal safety risk analysis is performed on the mutation points of the optical flow trajectory to obtain abnormal risk data of the dynamic trajectory; Step S6: Based on the dynamic trajectory abnormal risk data, abnormal targets are located on the omni-directional monitoring perception topology map, and abnormal safety prevention and control decisions are made to generate a campus safety decision engine.

2. The park security monitoring method based on dynamic visual recognition according to claim 1 is characterized in that: The specific steps of step S1 are: Step S11: Acquire multiple park monitoring videos based on park monitoring equipment; Step S12: performing video background texture feature analysis on multiple park surveillance videos to extract background texture features; Step S13: performing background noise analysis on the background texture features to obtain video background noise data; Step S14: performing multi-frequency filtering optimization on multiple park surveillance videos based on video background noise data to generate filtered noise-reduced surveillance videos; Step S15: performing three-dimensional illumination migration rendering on the filtered and denoised surveillance video to construct an illumination rendering surveillance video; Step S16: Perform low-frame smoothing on the illumination rendering monitoring video to generate a time-series monitoring image frame sequence.

3. The park security monitoring method based on dynamic visual recognition according to claim 2 is characterized in that: The specific steps of step S15 are: Extract multiple perspective images based on filtered denoising surveillance video; Identify brightness feature points of multiple viewing angle images and mark multiple viewing angle brightness feature points; Calculate the brightness of multiple viewing angle brightness feature points at the current timestamp to generate the brightness value of the image feature points; Identify brightness changes of image feature point brightness values ​​to obtain multi-view feature point brightness change data; Performing feature point spatial matching on multiple viewing angle brightness feature points to obtain a brightness feature point spatial matching relationship; Based on the spatial matching relationship of brightness feature points, the illumination intensity distribution data of multi-view feature points is analyzed to obtain the image illumination intensity distribution data; Perform three-dimensional illumination migration simulation on image illumination intensity distribution data to construct a dynamic three-dimensional illumination field; The park surveillance video is reconstructed by three-dimensional illumination field rendering according to the dynamic three-dimensional illumination field to construct the illumination rendering surveillance video.

4. The park security monitoring method based on dynamic visual recognition according to claim 2 is characterized in that: The specific steps of step S16 are: Define the sliding time window length; Performing time window division processing on the illumination rendering surveillance video according to the length of the sliding time window, and extracting surveillance videos of multiple time windows; Perform dynamic frame sliding analysis on each frame of the surveillance image based on the time interval of all image frames to identify images in low frame rate areas; Perform convolution learning prediction on low frame rate area images to extract video context convolution information; Based on the video context convolution information, low-frame smooth completion is performed on the low-frame rate area image to generate a smooth frame completion video stream; Based on the smooth frame completion video stream, frame-by-frame extraction is performed to obtain multiple monitoring image frames; Calculate the timestamp of each surveillance image frame; Performing equal-interval encoding based on the timestamp to obtain a number for each monitoring image frame; Based on the serial number of each monitoring image frame, multiple video image frames are serialized in a time sequence standard to generate a time sequence monitoring image frame sequence.

5. The park security monitoring method based on dynamic visual recognition according to claim 1 is characterized in that: The specific steps of step S2 are: Step S21: performing spatial location calculation on the park monitoring equipment to obtain coordinates of multiple equipment location points; Step S22: performing spatial distribution analysis on the coordinates of multiple device location points to generate monitoring device spatial distribution data; Step S23: performing device space distance calculation on the coordinates of multiple device location points to obtain distance parameters between devices; Step S24: Calculate the visual detection range of multiple park monitoring videos to obtain the device monitoring visual area; Step S25: performing cross-monitoring topological area analysis on the device monitoring visual area based on the distance parameter between devices, thereby generating inter-device monitoring topological association data; Step S26: Based on the monitoring topology association data between devices, the monitoring device spatial distribution data is reconstructed into a topology association network to construct an all-round monitoring perception topology map.

6. The park security monitoring method based on dynamic visual recognition according to claim 1 is characterized in that: The specific steps of step S3 are: Step S31: Perform visual detection of dynamic objects on the time-series monitoring image frame sequence and mark dynamic targets; Step S32: performing object type semantic recognition on the dynamic target to generate dynamic object types, where the dynamic object types include vehicles and pedestrians; Step S33: performing image segmentation of vehicles and pedestrians, and extracting vehicle target image frames and pedestrian target image frames; Step S34: performing frame-by-frame optical flow tracking analysis on the time-series monitoring image frame sequence according to the vehicle target image frame and the pedestrian target image frame to obtain the displacement vector of the dynamic image frame; Step S35: performing continuous frame displacement vector aggregation on the displacement vector of the dynamic image frame to generate a dynamic target optical flow trajectory.

7. The park security monitoring method based on dynamic visual recognition according to claim 1 is characterized in that: The specific steps of step S4 are: Step S41: sampling the dynamic target optical flow trajectory at multiple time points to obtain multiple optical flow trajectory sampling points; Step S42: extracting the timestamp of the optical flow trajectory sampling point; Step S43: calculating the interval displacement speed of multiple optical flow trajectory sampling points based on the timestamp, so as to obtain the displacement speed between the sampling points; Step S44: performing multi-point speed change calculation on the displacement speed between sampling points to generate the speed change rate of the optical flow trajectory; Step S45: identifying mutation points of multiple optical flow trajectory sampling points based on the optical flow trajectory speed change rate, thereby generating optical flow trajectory mutation points.

8. The park security monitoring method based on dynamic visual recognition according to claim 1 is characterized in that: The specific steps of step S5 are: Step S51: performing scene static building recognition on the time-series monitoring image frame sequence, and marking the park scene static building data; Step S52: Performing scene layout structure analysis on the park scene static building data to generate park scene static structure data; Step S53: performing geometric distance calculation on the mutation point of the optical flow trajectory based on the static scene structure data of the park, and identifying the distance between the mutation point and the scene structure; Step S54: performing personalized behavior pattern analysis according to the distance between the mutation point and the scene structure to generate dynamic trajectory behavior pattern features; Step S55: performing abnormal safety risk analysis on the dynamic trajectory behavior pattern characteristics, thereby obtaining dynamic trajectory abnormal risk data.

9. The park security monitoring method based on dynamic visual recognition according to claim 1 is characterized in that: The specific steps of step S6 are: Step S61: based on the dynamic trajectory abnormal risk data, reversely trace the dynamic target optical flow trajectory and mark the abnormal dynamic target; Step S62: quickly locate the target on the omni-directional monitoring perception topology map according to the abnormal dynamic target, so as to obtain the position coordinates of the abnormal target; Step S63: Perform intelligent warning response based on the position coordinates of the abnormal target and generate an intelligent warning signal; Step S64: Make abnormal safety prevention and control decisions based on the intelligent early warning signal to generate a campus safety decision engine.

10. A park security monitoring system based on dynamic visual recognition, characterized in that: The method for executing the park security monitoring method based on dynamic visual recognition as claimed in claim 1 comprises: The video enhancement module is used to obtain multiple park monitoring videos based on the park monitoring equipment; perform three-dimensional lighting migration rendering on multiple park monitoring videos, and perform low-frame smoothing to generate a time-series monitoring image frame sequence; The monitoring perception topology module is used to calculate the spatial location of the park monitoring equipment, reconstruct the topological association network, and build a comprehensive monitoring perception topology map; The optical flow tracking module is used to perform visual detection of dynamic objects on the time-series monitoring image frame sequence and perform frame-by-frame optical flow tracking analysis to generate the optical flow trajectory of the dynamic target; The optical flow trajectory analysis portal is used to sample the optical flow trajectory of dynamic targets at multiple time points, and then identify the mutation points to generate the optical flow trajectory mutation points; The abnormal safety risk module is used to identify static buildings in the scene of the time-series monitoring image frame sequence and generate static scene structure data of the park; based on the static scene structure data of the park, the abnormal safety risk analysis of the optical flow trajectory mutation points is performed to obtain the dynamic trajectory abnormal risk data; The security decision-making module is used to locate abnormal targets in the all-round monitoring perception topology map based on dynamic trajectory abnormal risk data, and make abnormal security prevention and control decisions to generate a campus security decision-making engine.

Images