Multi-camera linkage tracking personnel trajectory restoration intelligent security control system
Patent Information
- Application Number
- CN202611187202.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-06
- Publication Date
- 2026-10-16
Smart Images
Figure CN122934070A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of security control technology, and in particular to a smart security control system for reconstructing personnel trajectories through multi-camera linkage tracking. Background Technology
[0002] Park security systems are crucial infrastructure for ensuring the safety of various parks, including industrial parks, science parks, and logistics parks. With the development of artificial intelligence and computer vision technologies, park security systems are evolving from traditional passive monitoring towards intelligent and proactive approaches. Current smart park security systems typically deploy a large number of surveillance cameras, using technologies such as facial recognition and behavioral analysis to manage personnel within the park.
[0003] However, existing park security systems still have many technical shortcomings: First, each camera operates independently, lacking an effective linkage mechanism. When a target moves from the monitoring area of one camera to the monitoring area of another, the target is easily lost or its identity is confused. Although some cross-camera tracking technologies have been proposed, they mostly rely on target re-identification (Re-ID) technology for cross-camera matching. In complex scenarios such as dense crowds, changing lighting, and occlusion, the matching accuracy drops significantly, and it is difficult to continuously reconstruct the target's trajectory throughout the entire park.
[0004] Second, most existing camera linkage systems adopt "abnormal event triggered" linkage, that is, multi-camera collaborative tracking is only started after an abnormal event is detected. This is a passive response mode, which cannot establish a continuous full-domain tracking link when the target personnel enter the park. It lacks the ability to proactively perceive potential security risks that have not yet manifested as abnormal events.
[0005] Third, existing technologies for trajectory reconstruction mostly rely solely on visual image data for target matching and localization, lacking the fusion and utilization of multi-source heterogeneous sensor data. In situations such as nighttime, inclement weather, or when cameras have blind spots, the continuity and accuracy of trajectory reconstruction are difficult to guarantee. Furthermore, existing technologies lack the flexibility and adaptability of camera scheduling, making it difficult to dynamically optimize the camera's tracking angle based on the real-time movement of the target person. Summary of the Invention
[0006] Given that the existing systems still have problems such as each camera working independently and lacking an effective linkage mechanism, and that most existing camera linkage systems adopt "abnormal event triggered" linkage, that is, multi-camera collaborative tracking is only started after an abnormal event is detected, which is a passive response mode, and that existing technologies mostly rely on visual image data for target matching and positioning when realizing trajectory reconstruction, lacking the fusion and utilization of multi-source heterogeneous perception data, this invention is proposed.
[0007] Therefore, the purpose of this invention is to provide a smart security management system for personnel trajectory reconstruction based on multi-camera linkage tracking. Its purpose is to establish an active tracking link as soon as personnel enter the park, and to dynamically optimize the camera scheduling scheme according to the real-time movement status of the target through the camera linkage scheduling module, thereby realizing active and continuous tracking of all personnel in the park from entry to exit, filling the gap in existing technology in active full-area tracking.
[0008] To address the aforementioned technical problems, this invention provides the following technical solution: a smart security management system for personnel trajectory reconstruction based on multi-camera linkage tracking, comprising a front-end perception layer, a data transmission layer, an intelligent analysis layer, and a business application layer.
[0009] The front-end perception layer includes multiple smart cameras, UWB positioning base stations, access control card readers, and environmental sensors deployed in various areas of the park.
[0010] The data transmission layer is used to transmit the multimodal perception data collected by the front-end perception layer to the intelligent analysis layer.
[0011] The intelligent analysis layer includes a target detection and identity recognition module, which is used to perform real-time target detection on the video streams collected by the multi-channel intelligent cameras, identify personnel targets in the video images, and extract personnel identity features.
[0012] The multi-target cross-camera tracking module is used to continuously track and maintain the identity consistency of the same person target in cross-camera scenarios based on the output of the target detection and identity recognition module, combined with the spatial topology and field of view coverage of each camera, and generate segmented trajectory data of each person target.
[0013] The trajectory stitching and restoration module is used to combine the segmented trajectory data generated by the multi-target cross-camera tracking module with UWB positioning data, access control records and environmental sensor data to perform multi-source fusion and trajectory stitching, thereby generating a complete and continuous trajectory of personnel within the park.
[0014] The camera linkage scheduling module is used to dynamically calculate the optimal camera scheduling scheme based on the real-time location, movement direction and speed of the target person, combined with the pose parameters and field of view of each camera, and generate gimbal control commands to achieve optimal angle tracking of the target.
[0015] The abnormal behavior analysis module is used to perform multi-dimensional abnormal analysis of personnel behavior based on the complete and continuous trajectory and the preset abnormal behavior rule library, and generate abnormal early warning signals.
[0016] The business application layer includes a trajectory visualization display unit and an alarm linkage unit.
[0017] As a preferred embodiment of the intelligent security management system for personnel trajectory reconstruction based on multi-camera linkage tracking described in this invention, the multi-target cross-camera tracking module includes a target feature extraction submodule, which is used to extract the appearance feature vector and motion feature vector of personnel targets in video frames through a deep convolutional neural network.
[0018] The cross-camera target matching submodule is used to construct a cross-camera target matching model based on the similarity measure of the apparent feature vector, the continuity constraint of the movement direction, and the spatiotemporal consistency constraint when the target person moves out of the field of view of the first camera and into the field of view of the second camera, to determine whether the target in the first camera and the target in the second camera are the same person.
[0019] The identity consistency maintenance submodule is used to assign a globally unique identity to each identified person target and maintain the continuity of the identity after a successful cross-camera match.
[0020] As a preferred embodiment of the intelligent security management system for personnel trajectory reconstruction based on multi-camera linkage tracking described in this invention, the similarity measurement in the cross-camera target matching submodule adopts a weighted fusion strategy, specifically: calculating the cosine similarity between the apparent feature vector of the target person at the last moment in the field of view of the first camera and the apparent feature vector at the initial moment in the field of view of the second camera; calculating the directional consistency score between the movement direction of the target person at the end of the field of view of the first camera and the movement direction at the beginning of the field of view of the second camera; calculating the spatiotemporal constraint score based on the spatial positional relationship between the first camera and the second camera, the time when the target person leaves the field of view of the first camera and the time when they enter the field of view of the second camera; and weighting and summing the cosine similarity, directional consistency score, and spatiotemporal constraint score to obtain a comprehensive matching score. When the comprehensive matching score exceeds a preset threshold, the target person is determined to be the same person.
[0021] As a preferred embodiment of the intelligent security management system for personnel trajectory restoration based on multi-camera linkage tracking described in this invention, the trajectory stitching and restoration module includes a data spatiotemporal alignment submodule, which is used to align video trajectory data, UWB positioning data, and access control records according to a unified timestamp and spatial coordinate system.
[0022] The trajectory segment association submodule is used to construct an association map between trajectory segments based on spatiotemporally aligned multi-source data, and to achieve correct connection of different trajectory segments of the same person by solving the optimal association path.
[0023] The missing trajectory completion submodule is used to complete the missing trajectory segment in the event of a camera blind spot or loss of positioning signal, based on the target person's motion model, surrounding environmental constraints, and valid trajectory points before and after, through interpolation and prediction algorithms.
[0024] The 3D trajectory generation submodule is used to map the stitched and completed trajectory onto the 3D spatial model of the park, generating a 3D trajectory with time labels.
[0025] As a preferred embodiment of the intelligent security management system for personnel trajectory restoration through multi-camera linkage tracking described in this invention, the camera linkage scheduling module includes a target motion state estimation submodule, which is used to estimate the instantaneous velocity, motion direction and acceleration of the target based on the position changes of the target in multiple consecutive video images.
[0026] The camera coverage analysis submodule is used to calculate the coverage quality score of each camera for the current and predicted positions of the target based on the installation position, orientation angle, field of view, and zoom range of each camera.
[0027] The scheduling strategy generation submodule is used to generate an optimal camera scheduling strategy based on the coverage quality score, with the goal of maximizing tracking quality and minimizing camera action cost, using a multi-objective optimization algorithm. This strategy includes a list of cameras to be activated, the target PTZ parameters of each camera, and the target viewpoint switching sequence.
[0028] The gimbal coordination control submodule is used to convert the scheduling strategy into gimbal control commands for each camera and send them out for execution, so as to realize the smooth and coordinated tracking of the target by multiple cameras.
[0029] As a preferred embodiment of the intelligent security management system for personnel trajectory reconstruction based on multi-camera linkage tracking described in this invention, the abnormal behavior judgment module includes a trajectory feature extraction submodule, used to extract trajectory features from the complete continuous trajectory, wherein the trajectory features include movement speed, dwell time, path deviation, area access sequence, and trajectory curvature.
[0030] The multidimensional rule matching submodule is used to match the trajectory features with rules in a preset abnormal behavior rule library, which includes area intrusion rules, abnormal loitering rules, unauthorized path rules, speed anomaly rules, and time sequence anomaly rules.
[0031] The comprehensive risk scoring submodule is used to calculate the comprehensive risk score of personnel using a weighted comprehensive scoring method based on the matching results of each rule.
[0032] The early warning classification output submodule is used to output an abnormal early warning signal of the corresponding level based on the threshold range into which the comprehensive risk score falls.
[0033] As a preferred embodiment of the present invention, a smart security management method for reconstructing personnel trajectories through multi-camera linkage tracking is characterized by the following steps: S1, collecting real-time video streams through multiple smart cameras deployed in various areas of the park, collecting personnel location signals through UWB positioning base stations, collecting personnel access records through access control card readers, and collecting environmental parameter data through environmental sensors.
[0034] S2. Perform real-time target detection on the video streams collected by the multi-channel smart cameras, identify personnel targets in the video images and extract personnel identity features, and assign a temporary identity identifier to each detected personnel target.
[0035] S3. Based on the spatial topology and field of view coverage of each camera, the same person target is continuously tracked in cross-camera scenarios. The identity consistency across cameras is maintained through target appearance feature matching and motion consistency constraints, and segmented trajectory data of each person target is generated.
[0036] S4. The segmented trajectory data is spatiotemporally aligned and fused with UWB positioning data, access control records and environmental sensor data. By associating trajectory segments and completing missing trajectories, a complete and continuous three-dimensional trajectory of personnel within the park is generated.
[0037] S5. Based on the real-time location, direction of movement, and speed of the target person, and combined with the pose parameters and field of view of each camera, dynamically calculate the optimal camera scheduling scheme and generate gimbal control commands to achieve optimal angle collaborative tracking of the target by multiple cameras.
[0038] S6. Based on the complete and continuous trajectory and the preset abnormal behavior rule library, perform multi-dimensional abnormal judgment on personnel behavior and generate abnormal early warning signals.
[0039] S7. Visualize the complete continuous trajectory on the 3D map of the park and trigger the corresponding alarm linkage operation according to the level of the abnormal warning signal.
[0040] In a preferred embodiment of the present invention, in step S4, the missing trajectory completion specifically involves: when the duration of the lost positioning signal is less than a first preset threshold, short-term trajectory prediction is performed based on the historical motion state of the target using Kalman filtering; when the duration of the lost positioning signal is greater than or equal to the first preset threshold, the missing trajectory segment is completed using a bidirectional interpolation method, taking into account the target's last known location, the spatial layout constraints of the park, and the target's destination information.
[0041] In a preferred embodiment of the present invention, in step S5, the camera scheduling strategy is generated with the goal of maximizing tracking quality and minimizing camera action cost. The tracking quality includes the degree to which the target is centered in the camera's field of view, the proportion of the field of view occupied by the target, and the image clarity. The camera action cost includes the gimbal rotation angle, the zoom adjustment range, and the number of activated cameras.
[0042] In a preferred embodiment of the present invention, step S6, the abnormal behavior assessment specifically includes: extracting trajectory features from the complete continuous trajectory, the trajectory features including movement speed, dwell time, path deviation, area access sequence, and trajectory curvature; matching the trajectory features with rules in a preset abnormal behavior rule base, the abnormal behavior rule base including area intrusion rules, abnormal loitering rules, unauthorized path rules, speed anomaly rules, and time sequence anomaly rules; calculating the personnel's comprehensive risk score using a weighted comprehensive scoring method based on the matching results of each rule; and outputting an abnormal warning signal of the corresponding level based on the threshold range into which the comprehensive risk score falls.
[0043] Compared with the prior art, the present invention has at least the following beneficial effects: 1. The present invention establishes an active tracking link as soon as personnel enter the park, and dynamically optimizes the camera scheduling scheme according to the real-time movement status of the target through the camera linkage scheduling module, thereby realizing active and continuous tracking of all personnel in the park from entering to leaving the park, filling the gap in active full-domain tracking in the prior art.
[0044] 2. This invention uses a multi-dimensional matching strategy that integrates appearance feature similarity, motion direction consistency constraints, and spatiotemporal constraints to perform cross-camera target matching. Compared with existing solutions that rely solely on target re-identification technology, this invention has higher matching accuracy and stronger robustness in complex scenarios such as dense crowds and occlusion.
[0045] 3. This invention achieves accurate restoration of complete and continuous three-dimensional trajectories of personnel within the park by fusing video trajectory data, UWB high-precision positioning data, access control records, and environmental sensor data through trajectory segment association and missing trajectory completion technology. This solves the problem of trajectory breakage in scenarios such as blind spots and nighttime using a single vision solution.
[0046] 4. This invention constructs a camera scheduling model with tracking quality and motion cost as dual optimization objectives. It dynamically calculates the optimal scheduling strategy based on the real-time motion state of the target and generates gimbal collaborative control commands, thereby realizing optimal perspective collaborative tracking of the target by multiple cameras. This minimizes the motion cost of the cameras and the system energy consumption while ensuring tracking quality. Attached Figure Description
[0047] Figure 1 This is a schematic diagram of the framework of the intelligent security management system for personnel trajectory reconstruction based on multi-camera linkage tracking, as described in this invention.
[0048] Figure 2 This is a flowchart illustrating the steps of the intelligent security management system for multi-camera linkage tracking and personnel trajectory reconstruction according to the present invention. Detailed Implementation
[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0050] Example 1: System Architecture.
[0051] like Figure 1 and 2 As shown, this embodiment provides a smart security management system for personnel trajectory reconstruction through multi-camera linkage tracking, including a front-end perception layer, a data transmission layer, an intelligent analysis layer, and a business application layer.
[0052] The front-end perception layer includes multiple smart cameras deployed throughout the park, UWB positioning base stations, access control card readers, and environmental sensors. The smart cameras are distributed at key locations such as park entrances and exits, main roads, building entrances, elevator lobbies, and corridors. Each camera has pan-tilt-zoom (PTZ) control capabilities, supporting horizontal rotation, vertical rotation, and zoom operations. UWB positioning base stations are deployed in indoor and outdoor areas of the park to collect high-precision location signals from personnel wearing UWB tags. Access control card readers are installed at the entrances of various buildings and areas to collect personnel card swipe records. Environmental sensors, including light sensors and temperature and humidity sensors, collect environmental parameter data for camera scheduling decisions.
[0053] The data transmission layer is used to transmit multimodal sensing data collected by the front-end sensing layer to the intelligent analysis layer via wired network or 5G wireless network.
[0054] The intelligent analysis layer includes a target detection and identity recognition module, a multi-target cross-camera tracking module, a trajectory stitching and reconstruction module, a camera linkage scheduling module, and an abnormal behavior analysis module.
[0055] The target detection and identity recognition module performs real-time target detection on the video stream, uses a target detection algorithm based on deep convolutional neural networks to identify human targets, and extracts identity feature information such as facial features, body features, and clothing color features.
[0056] The multi-target cross-camera tracking module continuously tracks and maintains the identity consistency of the same person target in cross-camera scenarios based on the spatial topology and field of view coverage of each camera. This module specifically includes a target feature extraction submodule, a cross-camera target matching submodule, and an identity consistency maintenance submodule.
[0057] The trajectory stitching and restoration module is used to perform multi-source fusion and trajectory stitching of segmented trajectory data with UWB positioning data, access control records, and environmental sensor data. This module specifically includes a data spatiotemporal alignment submodule, a trajectory segment association submodule, a missing trajectory completion submodule, and a 3D trajectory generation submodule.
[0058] The camera coordination and scheduling module dynamically calculates the optimal camera scheduling scheme and generates PTZ control commands based on the real-time location, direction of movement, and speed of the target person, combined with the pose parameters and field of view of each camera. This module specifically includes a target motion state estimation submodule, a camera coverage analysis submodule, a scheduling strategy generation submodule, and a PTZ collaborative control submodule.
[0059] The abnormal behavior analysis module is used to perform multi-dimensional anomaly analysis based on complete continuous trajectories and a pre-set abnormal behavior rule base. This module specifically includes a trajectory feature extraction submodule, a multi-dimensional rule matching submodule, a comprehensive risk scoring submodule, and an early warning classification output submodule.
[0060] The business application layer includes a trajectory visualization unit and an alarm linkage unit. The trajectory visualization unit dynamically displays complete and continuous trajectories on a 3D map of the park, supporting historical trajectory playback and query. The alarm linkage unit triggers audible and visual alarms, push notifications, and linkage with access control systems based on the anomaly warning level.
[0061] Example 2: Cross-camera target matching method.
[0062] This embodiment details the specific implementation method of cross-camera target matching.
[0063] When the target person moves out of the field of view of the first camera and into the field of view of the second camera, the cross-camera target matching submodule starts the matching process.
[0064] First, the target feature extraction submodule extracts the appearance feature vector of the target person at the last moment in the field of view of the first camera, and the appearance feature vector at the initial moment in the field of view of the second camera, using a deep convolutional neural network. The appearance feature vector includes the joint encoding of multi-dimensional information such as facial features, body posture features, and clothing color features.
[0065] Then, the cross-camera target matching submodule calculates the cosine similarity between two apparent feature vectors. Simultaneously, it calculates the directional consistency between the movement direction of the target person at the end of the first camera's field of view and the movement direction at the beginning of the second camera's field of view, measured by the cosine of the angle between the two directional vectors. Furthermore, based on the spatial relationship between the first and second cameras and the time difference between the target person leaving the first camera's field of view and entering the second camera's field of view, it calculates the spatiotemporal consistency. The smaller the time difference and the closer the spatial distance, the higher the spatiotemporal consistency.
[0066] Finally, the similarity of appearance features, the degree of directional consistency, and the degree of spatiotemporal consistency are weighted and summed to obtain the comprehensive matching score. Each weight coefficient can be preset and adjusted according to the actual scenario requirements. When the comprehensive matching score exceeds a preset threshold, it is determined to be the same person, and the identity consistency maintenance submodule maintains the continuity of the globally unique identity identifier for that person. If the preset threshold is not reached, the monitoring of the target's appearance in other cameras continues, or confirmation is achieved through manual verification.
[0067] Example 3: Multi-source data fusion and trajectory reconstruction method.
[0068] This embodiment details the specific implementation method of multi-source data fusion and trajectory reconstruction.
[0069] The data spatiotemporal alignment submodule first aligns video trajectory data, UWB positioning data, and access control records according to a unified timestamp and spatial coordinate system. All data timestamps are uniformly converted to absolute time in milliseconds, and all location information is uniformly transformed to the park's spatial coordinate system. Video trajectory data is converted from pixel coordinates to world coordinates using camera calibration parameters; UWB positioning data is directly converted to the park's spatial coordinate system; and access control records are marked with their positions based on the preset coordinates of the access control points in the park's spatial model.
[0070] The trajectory segment association submodule constructs an association graph between trajectory segments based on spatiotemporally aligned multi-source data. Each trajectory segment contains the target person's identity identifier, time period, spatial location sequence, and motion characteristics. By calculating the temporal proximity, spatial continuity, and consistency of identity characteristics between different trajectory segments, association weights are constructed based on these three dimensions. The optimal association path is then solved to correctly connect different trajectory segments of the same person. This can be achieved using dynamic programming or graph theory shortest path algorithms.
[0071] The missing trajectory completion submodule is activated when the camera's field of view is in a blind spot or the UWB positioning signal is lost. When the positioning signal loss duration is less than a first preset threshold, short-term trajectory prediction is performed based on the target's historical motion state using Kalman filtering, and the possible positions during the loss period are predicted by extrapolating the target's continuous position observations before the loss. When the positioning signal loss duration is greater than or equal to the first preset threshold, the missing trajectory segment is completed using a bidirectional interpolation method, combining the target's last known position, park spatial layout constraints, and the target's possible destination information. Under the constraint of the park's passable area, a smooth candidate trajectory that conforms to kinematic laws is generated. If multiple candidate trajectories exist, the trajectory with the highest spatiotemporal matching degree with the access control record is selected.
[0072] The 3D trajectory generation submodule maps the stitched and completed trajectory onto the park's 3D spatial model. The park's 3D spatial model is pre-constructed using 3D modeling technology and includes floor information, passage information, and spatial coordinates for each building. A coordinate transformation matrix converts the 2D trajectory points into 3D spatial coordinate points, generating a time-stamped 3D trajectory that fully represents the three-dimensional activity trajectory of people in multi-story buildings.
[0073] Example 4: Camera linkage scheduling method.
[0074] This embodiment details the specific implementation method of camera linkage scheduling.
[0075] The target motion state estimation submodule estimates the instantaneous velocity, direction of motion, and acceleration of the target based on the positional changes of the target in multiple consecutive video frames, using either optical flow or Kalman filtering. Optical flow estimates motion parameters by calculating the pixel motion field of the target region between adjacent frames, while Kalman filtering obtains the optimal motion state value by performing state estimation on continuous position observations.
[0076] The camera coverage analysis submodule calculates the coverage quality score for each camera's current and predicted target positions based on the camera's installation location, orientation angle, field of view, and current zoom level. The score considers three factors: the target's central position within the field of view (closer to the center, higher score); the proportion of the field of view occupied by the target (highest score when the image size is within a preset suitable range); and image sharpness, assessed by analyzing image contrast and edge strength (higher sharpness, higher score).
[0077] The scheduling strategy generation submodule aims to maximize tracking quality and minimize camera action costs, employing a multi-objective optimization algorithm to generate the optimal camera scheduling strategy. Tracking quality is the weighted sum of the coverage quality scores of all active cameras, and camera action costs are the weighted sum of the gimbal rotation angle change, zoom adjustment range, and the number of active cameras. The scheduling strategy includes a list of cameras to be activated, the target PTZ parameters for each camera, and the target viewpoint switching sequence.
[0078] The PTZ collaborative control submodule converts the scheduling strategy into PTZ control commands for each camera and issues them for execution. During target viewpoint switching, a smooth transition strategy is adopted. When the target is about to move out of the current main camera's field of view, the adjacent camera is pre-adjusted to the target's viewpoint, and the old and new cameras continue to track simultaneously until the new camera completely locks onto the target before releasing the old camera, thus achieving smooth collaborative tracking.
[0079] Example 5: Intelligent judgment method for abnormal behavior.
[0080] This embodiment details the specific implementation method of intelligent judgment of abnormal behavior.
[0081] The trajectory feature extraction submodule extracts multi-dimensional trajectory features from a complete and continuous trajectory. Movement speed features include instantaneous speed sequence, average speed, and maximum speed; dwell time features include dwell time and dwell frequency in each area; path deviation features are the degree of deviation between the actual walking path and the preset reasonable path; area visit sequence features include the order and pattern of personnel visiting each area; and trajectory curvature features include the degree of curvature of the trajectory and the frequency of direction changes.
[0082] The multi-dimensional rule matching submodule matches the extracted trajectory features with rules in a pre-defined abnormal behavior rule library. The abnormal behavior rule library includes: area intrusion rules to detect whether personnel enter unauthorized areas; abnormal loitering rules to detect whether personnel linger or loiter repeatedly in unnecessary areas; unauthorized path rules to detect whether walking paths significantly deviate from reasonable paths; abnormal speed rules to detect whether movement speeds far exceed or fall far below normal ranges; and abnormal time sequence rules to detect whether specific areas are accessed during abnormal periods.
[0083] The comprehensive risk scoring submodule calculates the comprehensive risk score using a weighted comprehensive scoring method based on the matching results of each rule. The matching score for each rule ranges from 0 to 1, and the weight of each rule is preset based on its importance and historical early warning accuracy.
[0084] The early warning grading output submodule outputs corresponding level of abnormality warning signals based on the threshold range that the comprehensive risk score falls into. Multiple threshold ranges correspond to levels such as normal, attention, warning, and emergency. Higher-level warnings trigger the alarm linkage unit to perform linked operations such as audible and visual alarms, notification pushes, and access control.
[0085] Example 6: System Workflow.
[0086] This embodiment details the overall workflow of the system.
[0087] After the system starts up, the front-end perception layer continuously collects multimodal perception data and uploads it to the intelligent analysis layer.
[0088] In step S1, multiple smart cameras acquire real-time video streams, UWB positioning base stations acquire personnel location signals, access control card readers acquire passage records, and environmental sensors acquire environmental parameter data. Video stream data is acquired continuously in frames, UWB positioning data is actively reported at a fixed frequency, access control records are acquired when a card is swiped, and environmental data is sampled at a fixed period.
[0089] In step S2, the target detection and identity recognition module performs real-time target detection on the video stream, identifies each person target and extracts identity feature information. The first target to appear is assigned a temporary identity label, and targets with existing labels maintain label consistency in consecutive frames.
[0090] In step S3, the multi-target cross-camera tracking module continuously tracks targets across cameras based on the spatial topology of each camera. When a target switches between cameras, a cross-camera target matching process is initiated. The matching is determined by a combination of factors, including appearance feature similarity, motion direction consistency, and spatiotemporal consistency. Upon successful matching, a globally unique identifier is maintained, and segmented trajectory data for each target is generated.
[0091] In step S4, the trajectory stitching and restoration module performs spatiotemporal alignment and fusion of segmented trajectory data with UWB positioning data, access control records and environmental sensor data. By correctly connecting different trajectory segments of the same person through trajectory segment association, missing data periods are filled in, and finally a complete and continuous three-dimensional trajectory is generated.
[0092] Meanwhile, in step S5, the camera linkage scheduling module dynamically calculates the optimal scheduling scheme based on the target's real-time position, direction of movement, and speed, and generates gimbal control commands to achieve optimal perspective collaborative tracking of the target by multiple cameras. This process continues after personnel enter the park.
[0093] In step S6, the abnormal behavior analysis module performs multi-dimensional abnormal analysis based on the complete continuous trajectory and the preset abnormal behavior rule base. It extracts features such as speed, dwell time, path deviation, area access sequence and trajectory curvature, and calculates a comprehensive risk score after matching with the rule base and outputs a corresponding level of early warning signal.
[0094] In step S7, the trajectory visualization display unit will dynamically display the complete and continuous trajectory on the 3D map of the park in real time, supporting the playback and tracing of historical trajectories, and the alarm linkage unit will trigger corresponding linkage operations according to the warning level.
[0095] Thus, the system has completed a closed-loop management process from data collection, target tracking, trajectory reconstruction, camera linkage to anomaly analysis and alarm, achieving full-area, all-time, and all-dimensional security management of personnel within the park.
[0096] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. Smart security management and control system for personnel trajectory restoration through multi-camera collaborative tracking, characterized in that, It includes a front-end perception layer, a data transmission layer, an intelligent analysis layer, and a business application layer; The front-end perception layer includes multiple smart cameras, UWB positioning base stations, access control card readers, and environmental sensors deployed in various areas of the park. The data transmission layer is used to transmit the multimodal perception data collected by the front-end perception layer to the intelligent analysis layer; The intelligent analysis layer includes: The target detection and identity recognition module is used to perform real-time target detection on the video streams collected by the multi-channel smart cameras, identify personnel targets in the video images and extract personnel identity features; The multi-target cross-camera tracking module is used to continuously track and maintain the identity consistency of the same person target in cross-camera scenarios based on the output of the target detection and identity recognition module, combined with the spatial topology and field of view coverage of each camera, and generate segmented trajectory data of each person target. The trajectory stitching and restoration module is used to combine the segmented trajectory data generated by the multi-target cross-camera tracking module with UWB positioning data, access control records and environmental sensor data to perform multi-source fusion and trajectory stitching, and generate a complete and continuous trajectory of personnel in the park. The camera linkage scheduling module is used to dynamically calculate the optimal camera scheduling scheme based on the real-time location, movement direction and speed of the target person, combined with the pose parameters and field of view of each camera, and generate gimbal control commands to achieve optimal angle tracking of the target. The abnormal behavior analysis module is used to perform multi-dimensional abnormal analysis of personnel behavior based on the complete and continuous trajectory and the preset abnormal behavior rule library, and generate abnormal early warning signals. The business application layer includes a trajectory visualization display unit and an alarm linkage unit.
2. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 1, characterized in that: The multi-target cross-camera tracking module includes: The target feature extraction submodule is used to extract the appearance feature vector and motion feature vector of people in video frames through a deep convolutional neural network; The cross-camera target matching submodule is used to construct a cross-camera target matching model based on the similarity measure of the apparent feature vector, the continuity constraint of the movement direction, and the spatiotemporal consistency constraint when the target person moves out of the field of view of the first camera and enters the field of view of the second camera, to determine whether the target in the first camera and the target in the second camera are the same person. The identity consistency maintenance submodule is used to assign a globally unique identity to each identified person target and maintain the continuity of the identity after a successful cross-camera match.
3. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 2, characterized in that: The similarity measurement in the cross-camera target matching submodule adopts a weighted fusion strategy, specifically: calculating the cosine similarity between the apparent feature vector of the target person at the last moment in the field of view of the first camera and the apparent feature vector at the initial moment in the field of view of the second camera. Calculate the directional consistency score between the direction of movement of the target person at the end of the field of view of the first camera and the direction of movement at the beginning of the field of view of the second camera; Based on the spatial relationship between the first and second cameras, the time when the target person leaves the field of view of the first camera and the time when they enter the field of view of the second camera, the spatiotemporal constraint score is calculated. The cosine similarity, directional consistency score, and spatiotemporal constraint score are weighted and summed to obtain a comprehensive matching score. When the comprehensive matching score exceeds a preset threshold, the individuals are determined to be the same person.
4. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 3, characterized in that: The trajectory stitching and restoration module includes: The data spatiotemporal alignment submodule is used to align video trajectory data, UWB positioning data, and access control records according to a unified timestamp and spatial coordinate system. The trajectory segment association submodule is used to construct an association map between trajectory segments based on spatiotemporally aligned multi-source data, and to achieve correct connection of different trajectory segments of the same person by solving the optimal association path; The missing trajectory completion submodule is used to complete the missing trajectory segment in the case of camera blind spot or loss of positioning signal, based on the target person's motion model, surrounding environmental constraints and effective trajectory points before and after, through interpolation and prediction algorithms. The 3D trajectory generation submodule is used to map the stitched and completed trajectory onto the 3D spatial model of the park, generating a 3D trajectory with time labels.
5. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 4, characterized in that: The camera linkage scheduling module includes: The target motion state estimation submodule is used to estimate the instantaneous velocity, direction of motion, and acceleration of the target based on the position changes of the target in multiple consecutive video images; The camera coverage analysis submodule is used to calculate the coverage quality score of each camera for the current and predicted positions of the target based on the installation position, orientation angle, field of view and zoom range of each camera. The scheduling strategy generation submodule is used to generate the optimal camera scheduling strategy based on the coverage quality score, with the goal of maximizing tracking quality and minimizing camera action cost, using a multi-objective optimization algorithm. The strategy includes a list of cameras to be activated, the target PTZ parameters of each camera, and the target view switching sequence. The gimbal coordination control submodule is used to convert the scheduling strategy into gimbal control commands for each camera and send them out for execution, so as to realize the smooth and coordinated tracking of the target by multiple cameras.
6. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 5, characterized in that: The abnormal behavior analysis module includes: The trajectory feature extraction submodule is used to extract trajectory features from the complete continuous trajectory, the trajectory features including movement speed, dwell time, path deviation, area visit sequence and trajectory curvature; The multidimensional rule matching submodule is used to match the trajectory features with rules in a preset abnormal behavior rule library, which includes area intrusion rules, abnormal loitering rules, unauthorized path rules, speed anomaly rules, and time sequence anomaly rules. The comprehensive risk scoring submodule is used to calculate the comprehensive risk score of personnel based on the matching results of each rule using a weighted comprehensive scoring method. The early warning classification output submodule is used to output an abnormal early warning signal of the corresponding level based on the threshold range into which the comprehensive risk score falls.
7. A smart security management and control method for personnel trajectory restoration through multi-camera coordinated tracking, applied to the system according to any one of claims 1 to 6, characterized in that, Includes the following steps: S1. Real-time video streams are collected through multiple smart cameras deployed in various areas of the park, personnel location signals are collected through UWB positioning base stations, personnel access records are collected through access control card readers, and environmental parameter data are collected through environmental sensors. S2. Perform real-time target detection on the video streams collected by the multi-channel smart cameras, identify personnel targets in the video frame and extract personnel identity features, and assign a temporary identity identifier to each detected personnel target; S3. Based on the spatial topology and field of view coverage of each camera, the same person target is continuously tracked in cross-camera scenarios. The identity consistency across cameras is maintained through target appearance feature matching and motion consistency constraints, and segmented trajectory data of each person target is generated. S4. The segmented trajectory data is spatiotemporally aligned and fused with UWB positioning data, access control records and environmental sensor data. By associating trajectory segments and completing missing trajectories, a complete and continuous three-dimensional trajectory of personnel in the park is generated. S5. Based on the real-time location, direction of movement, and speed of the target person, and combined with the pose parameters and field of view of each camera, dynamically calculate the optimal camera scheduling scheme and generate gimbal control commands to achieve optimal angle collaborative tracking of the target by multiple cameras. S6. Based on the complete and continuous trajectory and the preset abnormal behavior rule library, perform multi-dimensional abnormal analysis of personnel behavior and generate abnormal early warning signals; S7. Visualize the complete continuous trajectory on the 3D map of the park and trigger the corresponding alarm linkage operation according to the level of the abnormal warning signal.
8. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 7, characterized in that: In step S4, missing trajectory completion specifically involves: when the location signal loss time is less than a first preset threshold, short-term trajectory prediction is performed based on the target's historical motion state using Kalman filtering. When the duration of the lost positioning signal is greater than or equal to the first preset threshold, a bidirectional interpolation method is used to complete the missing trajectory segment, taking into account the target's last known location, the spatial layout constraints of the park, and the target's destination information.
9. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 8, characterized in that: In step S5, the camera scheduling strategy is generated with the goal of maximizing tracking quality and minimizing camera action cost. The tracking quality includes the degree to which the target is centered in the camera's field of view, the proportion of the field of view occupied by the target, and the image clarity. The camera action cost includes the gimbal rotation angle, zoom adjustment range, and the number of activated cameras.
10. The intelligent security control system for personnel trajectory reconstruction based on multi-camera linkage tracking according to claim 9, characterized in that: In step S6, the abnormal behavior assessment specifically includes: extracting trajectory features from the complete continuous trajectory, the trajectory features including movement speed, dwell time, path deviation, area access sequence, and trajectory curvature; matching the trajectory features with rules in a preset abnormal behavior rule base, the abnormal behavior rule base including area intrusion rules, abnormal loitering rules, unauthorized path rules, speed anomaly rules, and time sequence anomaly rules; calculating the personnel's comprehensive risk score using a weighted comprehensive scoring method based on the matching results of each rule; and outputting an abnormal warning signal of the corresponding level based on the threshold range into which the comprehensive risk score falls.