Video twin park human and vehicle flow security analysis method and system based on live-action three-dimensional

Abstract

The invention belongs to the technical field of prediction analysis, and particularly relates to a video twinning park people and vehicle flow security analysis method and system based on live-action three-dimensional, and the system comprises a video twinning engine module, a video space correction and fusion module, an LI position intelligent analysis module, a real-time simulation calculation module and an intelligent analysis decision module. According to the invention, the video twin engine module synchronously integrates three types of heterogeneous data sources, a video monitoring system, an original video stream, an artificial intelligence analysis system, a target detection result, an Internet of Things sensing system and sensor data, so that dynamic mapping is realized in a unified three-dimensional scene model, a traditional system data island is broken through, and the real-time performance of the system is improved. A global space-time association view is constructed, Internet of Things data such as environment temperature and humidity and access control states can be overlaid on personnel trajectories, and the behavior analysis accuracy is improved; the positioning of the video shielding area can be complemented through Internet of Things equipment, such as a positioning label, and the influence of a monitoring blind area is eliminated.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of predictive analysis, and specifically relates to a video twin park personnel and vehicle flow security analysis method and system based on real three-dimensional. BACKGROUND

[0002] With the rapid development of Internet of Things, artificial intelligence and digital twin technology, intelligent monitoring systems play an increasingly important role in park safety management. As an important part of new generation information technology, digital twin technology constructs virtual mapping of physical entities through digital means, realizing real-time monitoring, simulation analysis and optimal control of the physical world. Video twin technology, as an innovative application of digital twin in video monitoring field, upgrades traditional two-dimensional video monitoring to a three-dimensional space-time perception system, providing a new technical path for park personnel flow management.

[0003] In the prior art, the park personnel monitoring system mainly includes the following technical solutions: traditional video monitoring system, intelligent video analysis system, digital twin visualization system and Internet of Things sensing system.

[0004] However, the above prior art solutions have the following main shortcomings:

[0005] Firstly, traditional video monitoring, intelligent analysis, digital twin and Internet of Things systems operate independently, lack effective data fusion mechanism, and cannot realize collaborative analysis and comprehensive utilization of multi-source data.

[0006] And traditional video monitoring can only provide two-dimensional picture information, and cannot accurately obtain the precise position coordinates of personnel in three-dimensional space, limiting the accuracy of spatial analysis.

[0007] And the existing digital twin system mostly uses static three-dimensional model, lacks dynamic fusion capability with real-time video data, and cannot realize real-time reflection of park state.

[0008] At the same time, the traditional system mainly relies on manual observation and simple rule judgment, lacks intelligent analysis capability based on space-time position, and cannot perform complex personnel flow pattern recognition and prediction.

[0009] Finally, two-dimensional video pictures cannot provide intuitive spatial relationship display, and operating personnel have difficulty in quickly understanding the overall situation and personnel distribution of the park. SUMMARY

[0010] To solve the above problems in the prior art, the present application provides a video twin park personnel and vehicle flow security analysis method and system based on real three-dimensional, aiming to solve the technical problems of data isolation, inaccurate spatial positioning, insufficient real-time performance and limited analysis capability in the park personnel monitoring system in the prior art.

[0011] In order to achieve the above object, the present application provides the following technical scheme: a video twin park people and vehicle flow security analysis system based on real three-dimensional video, comprising a video twin engine module, a video space correction and fusion module, an LI position intelligent analysis module, a real-time simulation calculation module and an intelligent analysis decision module; the video twin engine module is used for loading a three-dimensional scene model, fusing video monitoring system data, artificial intelligence analysis system data and Internet of Things sensing system data; the video space correction and fusion module applies a geometric transformation algorithm to correct the distortion of a two-dimensional video picture, and establishes a mapping relationship between two-dimensional pixel coordinates and three-dimensional world coordinates; the LI position intelligent analysis module identifies personnel targets in the video based on a target detection algorithm, and applies a three-dimensional coordinate conversion model to output personnel three-dimensional space position coordinates; the real-time simulation calculation module processes video data streams using a streaming calculation framework, and maintains a dynamic mapping relationship between physical personnel and virtual objects in a three-dimensional scene; and the intelligent analysis decision module is used for spatio-temporal correlation analysis of personnel trajectory data.

[0012] Further, the geometric transformation algorithm comprises a perspective transformation algorithm, an affine transformation algorithm and a radial distortion correction algorithm.

[0013] Further, the three-dimensional coordinate conversion model converts two-dimensional pixel coordinates of the detected personnel targets into three-dimensional world coordinates containing longitude, latitude and altitude based on an intrinsic matrix, an extrinsic matrix and ground constraint conditions of a camera.

[0014] Further, the streaming calculation framework of the real-time simulation calculation module has a response delay of less than or equal to 100 milliseconds.

[0015] Further, the spatio-temporal correlation analysis of the intelligent analysis decision module comprises a time series analysis algorithm and a spatial clustering algorithm.

[0016] A video twin park people and vehicle flow security analysis method based on real three-dimensional video, comprising the following steps: step one, acquiring multiple video stream data; step two, applying a geometric transformation algorithm to correct video distortion, and fusing the corrected video data to a three-dimensional digital twin scene based on the established mapping relationship; step three, identifying personnel targets in the video based on a target detection algorithm, and applying a three-dimensional coordinate conversion model to output personnel three-dimensional space position coordinates; and step four, using a streaming calculation framework to update the dynamic correlation between physical personnel and virtual objects in a three-dimensional scene in real time.

[0017] Further, the three-dimensional coordinate conversion model applies an intrinsic matrix, an extrinsic matrix and ground constraint conditions of a camera for coordinate conversion.

[0018] Further, the dynamic correlation update comprises applying a Kalman filtering algorithm and a Hungarian algorithm for multi-target tracking.

[0019] Further, the space-time correlation analysis includes extracting personnel moving speed parameters, direction parameters and stay time parameters.

[0020] Further, the space-time correlation analysis includes generating personnel flow trend prediction results based on historical trajectory data and real-time position data.

[0021] Compared with the prior art, the present application has the following beneficial effects:

[0022] 1. In the present application, the video twin engine module synchronously integrates three types of heterogeneous data sources, namely, a video monitoring system (original video stream), an artificial intelligence analysis system (target detection result) and an Internet of Things sensing system (sensor data), so as to realize dynamic mapping in a unified three-dimensional scene model, break the data island of traditional systems, construct a global space-time correlation view, superimpose Internet of Things data such as environment temperature and humidity and access control state on personnel trajectory, and improve the accuracy of behavior analysis; the positioning of video occlusion areas can be completed by Internet of Things devices (such as positioning tags), and the influence of monitoring blind areas is eliminated.

[0023] 2. In the prior art, video monitoring only provides two-dimensional pictures and cannot support space analysis. In the present application, the LI position intelligent analysis module extracts personnel pixel coordinates through a three-stage calculation, target detection algorithm (YOLO / R-CNN), solves three-dimensional world coordinates (longitude, latitude and altitude) based on camera internal parameter matrix, external parameter matrix and ground constraint, realizes centimeter-level accurate mapping between physical space and digital space, accurately calculates space parameters such as personnel density (number of people per unit area) and moving direction angle, and provides position basis for falling risk early warning of three-dimensional scenes such as elevated platforms and staircases.

[0024] 3. In the prior art, a digital twin system relies on a static model and cannot respond to changes in real time. In the present application, the real-time simulation calculation module adopts a two-level synchronous architecture. In the flow processing layer, an Apache Flink / Kafka stream engine processes 64 video streams with a delay of ≤100 ms. In the mapping layer, Kalman filtering predicts the next frame position of a target, the Hungarian algorithm solves cross-lens ID correlation, and the position, speed and direction of a three-dimensional virtual object are dynamically updated, so that the digital twin and the physical entity are strictly synchronized, thereby triggering virtual scene alarm immediately when an abnormal event such as crowd running occurs, and supporting dynamic simulation deduction of evacuation paths based on real-time positions. BRIEF DESCRIPTION OF DRAWINGS

[0025] The accompanying drawings are included to provide a further understanding of the application and constitute a part of the specification, illustrate the application and are used to explain the application together with the embodiments of the application, and do not constitute a limitation on the application. In the drawings:

[0026] Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0028] Embodiment 1:

[0029] Please refer to Figure 1 The embodiment provides the following technical solutions: a video twin park security analysis system based on real scene three dimensions, comprising a video twin engine module, a video space correction and fusion module, an LI position intelligent analysis module, a real-time simulation calculation module and an intelligent analysis decision module; the video twin engine module is used for loading a three-dimensional scene model, fusing video monitoring system data, artificial intelligence analysis system data and Internet of Things sensing system data; the video space correction and fusion module applies a geometric transformation algorithm to correct distortion of a two-dimensional video picture, and establishes a mapping relationship between two-dimensional pixel coordinates and three-dimensional world coordinates; the LI position intelligent analysis module identifies personnel targets in a video based on a target detection algorithm, and applies a three-dimensional coordinate conversion model to output personnel three-dimensional space position coordinates; the real-time simulation calculation module adopts a streaming calculation framework to process a video data stream, and maintains a dynamic mapping relationship between physical personnel and virtual objects in a three-dimensional scene; and the intelligent analysis decision module is used for spatio-temporal correlation analysis of personnel trajectory data.

[0030] Specifically, the video twin engine module loads a three-dimensional park scene model constructed by TB-level GIS data; 64 4K video streams (1920x1080), AI system target recognition results and IOT sensor data are synchronously fused. The video space correction and fusion module performs perspective distortion correction and radial distortion compensation on multi-view video streams, and establishes a homography transformation matrix from a pixel coordinate system to a three-dimensional world coordinate system. The LI position intelligent analysis module adopts a YOLOv5 model to detect personnel targets in a video in real time; and three-dimensional geodetic coordinates of the targets are solved based on camera calibration parameters. The real-time simulation calculation module deploys an Apache Flink streaming processing engine, and processes 64 video data streams in parallel; and physical target IDs are dynamically bound to three-dimensional virtual objects. The intelligent analysis decision module: performs DBSCAN spatial clustering and LSTM time series analysis on trajectory data.

[0031] Thus, the problems of isolation and non-intuitive visualization of multi-source data are solved, spatio-temporal correlation analysis of global personnel trajectories and environmental parameters is realized by fusing video, AI and IOT data to a three-dimensional scene through the video twin engine module, data utilization is improved by 80%, and decision efficiency is improved.

[0032] The geometric transformation algorithm includes a perspective transformation algorithm, an affine transformation algorithm, and a radial distortion correction algorithm.

[0033] Specifically, the perspective transformation algorithm calculates a projection transformation through a camera extrinsic matrix [R|t]; the affine transformation algorithm solves an affine matrix based on three groups of non-collinear feature points; the radial distortion correction algorithm compensates for lens distortion coefficients k1, k2, k3 by using a Brown-Conrady model.

[0034] Thus, the spatial distortion error of a traditional video monitoring is eliminated, the geometric consistency of a two-dimensional video and a three-dimensional scene is ensured through the perspective transformation, the affine transformation, and the radial distortion correction algorithm, and a basic condition is provided for centimeter-level positioning.

[0035] The three-dimensional coordinate conversion model converts the two-dimensional pixel coordinates of a detected personnel target into three-dimensional world coordinates containing longitude, latitude, and altitude based on an intrinsic matrix, an extrinsic matrix, and a ground constraint condition of a camera.

[0036] Specifically, the three-dimensional coordinate conversion model performs: inputting target pixel coordinates (u, v) and a camera intrinsic matrix K; combining an extrinsic matrix [R|t] to calculate the coordinates of the target in a camera coordinate system; introducing a ground constraint equation z=0 to solve three-dimensional world coordinates (x, y, z), and the precision error is ≤10 cm.

[0037] Thus, the limitation of two-dimensional positioning precision is broken through, and the three-dimensional coordinate conversion model of the intrinsic matrix, the extrinsic matrix, and the ground constraint is used to output the centimeter-level position coordinates (longitude, latitude, and altitude) containing the altitude, and the spatial positioning precision is obviously improved compared with the traditional scheme.

[0038] The response delay of the stream computing framework of the real-time simulation calculation module is less than or equal to 100 milliseconds.

[0039] Specifically, the stream computing framework uses a sliding window with a window size of 50 ms to process the video stream; a target mapping table is maintained through a state snapshot mechanism to ensure that the state update delay is ≤100 ms; and real-time concurrent calculation of 64 channels of 1080P video @ 30 fps is supported.

[0040] The problem of insufficient real-time performance of digital twins is solved, millisecond-level state synchronization of physical and virtual objects of 64 channels of video is realized through the stream computing framework and dynamic mapping maintenance, and the real-time monitoring demand of a large-scale park is met.

[0041] The spatiotemporal correlation analysis of the intelligent analysis and decision module includes a time series analysis algorithm and a spatial clustering algorithm.

[0042] Specifically, the time series analysis algorithm uses an LSTM network to predict the trajectory offset; the spatial clustering algorithm is based on DBSCAN density clustering with a Euclidean distance threshold ε=1.5 m;

[0043] Thus, the complex scene analysis capability is enhanced, and through a time sequence analysis algorithm and a space clustering algorithm, a space-time mode such as personnel gathering and dispersing is recognized, and the analysis accuracy is high.

[0044] The video twin park personnel and vehicle flow security analysis method based on a real three-dimensional scene comprises the following steps: step one, acquiring multiple video stream data; step two, correcting video distortion by applying a geometric transformation algorithm, and fusing the corrected video data to a three-dimensional digital twin scene based on the established mapping relationship; step three, recognizing personnel targets in the video based on a target detection algorithm, and applying a three-dimensional coordinate conversion model to output personnel three-dimensional space position coordinates; and step four, using a streaming computing framework to update the dynamic association between physical personnel and virtual objects in the three-dimensional scene in real time.

[0045] Specifically, the data acquisition step synchronously accesses 64 RTSP video streams and IOT device data; the space correction step performs radial distortion correction on each frame of image, with compensation coefficients k1=-0.21 and k2=0.03; the pixel coordinates are mapped to a three-dimensional scene through a homography matrix H; the positioning calculation step uses YOLOv5 to detect personnel target frames; the three-dimensional coordinates are calculated based on a camera pitch angle θ=35°±5°; the dynamic synchronization step updates the virtual object position / velocity vector at a period of 50 ms; and the trend analysis step extracts abnormal moving segments with a trajectory speed v>1.5 m / s.

[0046] Thus, the defects of data fusion and real-time analysis are solved, and through a method chain of video correction fusion, three-dimensional coordinate conversion, and streaming update, the three technical goals of multi-source data integration, centimeter-level positioning, and millisecond-level response are synchronously achieved.

[0047] The three-dimensional coordinate conversion model applies an intrinsic matrix, an extrinsic matrix, and a ground constraint condition of a camera to perform coordinate conversion.

[0048] Specifically, the three-dimensional coordinate conversion is realized through the following matrix operation:

[0049] [X,Y,Z] ∧T =R ∧ (-1)*(K ∧ (-1)*[u,v,1] ∧T *d-t),

[0050] where d is a measured distance of a target to a camera lens, R is a rotation matrix, and t is a translation vector.

[0051] Thus, the reliability of three-dimensional positioning results is ensured, and through the coordinate conversion method of the intrinsic matrix, the extrinsic matrix, and the ground constraint, the coordinate calculation error caused by the installation angle of the camera is eliminated.

[0052] The dynamic association update comprises applying a Kalman filtering algorithm and a Hungarian algorithm for multi-target tracking.

[0053] Specifically, the Kalman filter predicts the next frame position of the target, and the state vector is [x, y, vx, vy];The Hungarian algorithm matches the cross-frame target ID based on the cost matrix of IOU> 0.7;

[0054] Thus, the cross-lens target tracking continuity is maintained, the target identity jump problem in the occlusion scene is solved by the multi-target tracking method of Kalman filtering and Hungarian algorithm.

[0055] The space-time correlation analysis includes extracting the moving speed parameter, the direction parameter and the stay time parameter of the personnel.

[0056] Specifically, the moving speed is calculated based on the coordinate difference of adjacent frames, and the unit is m / s;The direction angle is the heading angle with the north as 0°, and the precision is ± 5 degrees;The stay time is the duration of the position change <0.2m, and the unit is second;

[0057] Thus, the behavior characteristics of the personnel are quantified, and the behavior baseline model can be calculated by extracting the moving speed, direction and stay time parameters.

[0058] The space-time correlation analysis includes generating the personnel flow trend prediction result based on the historical trajectory data and real-time position data.

[0059] Specifically, the personnel flow trend prediction includes inputting the historical 7-day trajectory data set of the same period;The ARIMA model is used to predict the personnel density distribution in the future 15 minutes;The high-risk gathering area coordinate list is outputted;

[0060] Thus, the flow trend prediction ability is improved, the personnel gathering area and path planning suggestion are outputted by machine learning prediction of historical trajectory and real-time data, and active safety management is supported.

[0061] Embodiment 2:

[0062] Please refer to Figure 1 In this embodiment, the staff deploys and uses in a certain R&D building according to the method disclosed in the present application;

[0063] The system adopts 64 Hikvision DS-2CD6987G2-LSU / SL 4K ultra-wide-angle cameras with a resolution of 1920x1080 to cover all the corridors and viewing platforms in the building, and combines 200 P-TDOA UWB positioning tags with an accuracy of ±30 cm, Honeywell HIH9000 temperature and humidity sensors, and Siemens SIWAREX WP321 access control sensors to build a multi-source perception layer. The computing platform is configured with 3 Dell PowerEdge R750 servers, each equipped with dual Intel Xeon Gold 6338 processors and 4 NVIDIA T4 GPUs for stream processing. The SuperMap GIS platform is equipped with an NVIDIA RTX A6000 graphics card to achieve three-dimensional rendering. During implementation, the system accesses 64 video streams through the RTSP protocol, synchronously receives 10Hz updated UWB data and 5-second interval sensor data, and uses the Brown-Conrady model for radial distortion correction with a correction coefficient k1=-0.21 and k2=0.03. Based on 108 ground control points, a homography matrix H is established to achieve an alignment error of ≤5 cm in the WGS-84 coordinate system. The LI position intelligent analysis module runs the YOLOv5s model on the NVIDIA T4 to achieve 68FPS detection, and combines the camera intrinsic matrix K, the extrinsic matrix R|t, and the height constraint z=3.2m to calculate the personnel position through the three-dimensional coordinate conversion formula:

[0064]

[0065] The actual positioning error is ≤10 cm. The real-time simulation calculation module processes data streams with a 50ms sliding window using Apache Flink, and a Kalman filter predicts the next frame position state vector x, y, vx, vy. The Hungarian algorithm correlates target IDs based on an IOU>0.7 across lenses, and automatically switches UWB data to maintain trajectory continuity when the video is blocked.

[0066] The intelligent analysis and decision-making module performs multi-dimensional monitoring, real-time calculation of stair area density exceeding 2 people / square meter to trigger an orange warning, detection of abnormal behavior with a moving speed >1.5m / s and a direction angle 10-second mutation >90 degrees, analysis of historical 7-day data using an ARIMA model to predict future 15-minute viewing platform crowd distribution, dynamic generation of evacuation paths, and red labeling of high-risk areas in the three-dimensional scene. On September 13, 2024, at 15:23, the system detected a sudden increase in density on the 5th floor rotating staircase to 3.1 people / square meter and a ground humidity of 85%RH, identified abnormal changes in the direction angle of 3 personnel, and predicted the trajectory of 1 person pointing to the guardrail gap using Kalman filtering. Within 82 milliseconds, a three-level alarm was triggered: highlighting the high-risk area in the three-dimensional scene, pushing instructions to the patrol robot, and unlocking the backup access control. Post-verification shows that the data utilization rate and decision-making efficiency are improved compared to traditional solutions.

[0067] The working principle of the present application: the video twin engine module loads the three-dimensional scene model, synchronously fuses the multi-channel video stream data of the video monitoring system, such as 64 channels of 4K video, 1920x1080, the target detection result of the artificial intelligence analysis system and the sensor data of the Internet of Things sensing system;

[0068] The video space correction and fusion module applies perspective transformation algorithm, affine transformation algorithm and radial distortion correction algorithm to correct video distortion, and establishes the mapping relationship between two-dimensional pixel coordinates and three-dimensional world coordinates;

[0069] The LI position intelligent analysis module identifies personnel targets in the video based on the target detection algorithm, and applies a three-dimensional coordinate conversion model to convert two-dimensional pixel coordinates (u, v) into three-dimensional world coordinates (x, y, z) containing longitude, latitude and altitude, with an accuracy error of ≤10 cm;

[0070] The real-time simulation calculation module uses a streaming computing framework to process video data streams, predicts the next frame position of the target through Kalman filtering and the Hungarian algorithm, and maintains the dynamic mapping of physical personnel and virtual objects in the three-dimensional scene with a delay of ≤100 ms;

[0071] The intelligent analysis and decision module performs spatio-temporal correlation analysis on the trajectory data, extracts personnel movement speed, direction and stay time parameters through time series analysis algorithm and spatial clustering algorithm, and generates personnel flow trend prediction results based on historical trajectory and real-time data.

[0072] Finally, it should be noted that: the above only describes the preferred embodiments of the present application and does not limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A video twin park people and vehicle flow security analysis system based on real three-dimensional, characterized in that: The video twin engine module, the video space correction and fusion module, the LI position intelligent analysis module, the real-time simulation calculation module and the intelligent analysis decision module are included. The video twin engine module is used for loading a three-dimensional scene model, fusing video monitoring system data, artificial intelligence analysis system data and Internet of Things sensing system data. The video space correction and fusion module applies a geometric transformation algorithm to correct the distortion of a two-dimensional video picture, and establishes a mapping relationship between two-dimensional pixel coordinates and three-dimensional world coordinates. The LI position intelligent analysis module identifies personnel targets in a video based on a target detection algorithm, and applies a three-dimensional coordinate conversion model to output personnel three-dimensional space position coordinates. The real-time simulation calculation module uses a streaming calculation framework to process video data streams, and maintains a dynamic mapping relationship between physical personnel and virtual objects in a three-dimensional scene. The intelligent analysis decision module is used for spatio-temporal correlation analysis of personnel trajectory data.

2. The real-scene three-dimensional-based video twin park human-vehicle flow security analysis system according to claim 1, characterized in that: The geometric transformation algorithm includes a perspective transformation algorithm, an affine transformation algorithm and a radial distortion correction algorithm. 3.The real-scene three-dimensional based video twin garden human-vehicle flow security analysis system according to claim 1, characterized in that: The three-dimensional coordinate conversion model converts two-dimensional pixel coordinates of detected personnel targets into three-dimensional world coordinates containing longitude, latitude and altitude based on an intrinsic matrix, an extrinsic matrix and ground constraints of a camera.

4. The real-scene three-dimensional based video twin park human-vehicle flow security analysis system according to claim 1, characterized in that: The streaming calculation framework of the real-time simulation calculation module has a response delay of less than or equal to 100 milliseconds.

5. The real-scene three-dimensional based video twin garden human-vehicle flow security analysis system according to claim 1, characterized in that: The spatio-temporal correlation analysis of the intelligent analysis decision module includes a time series analysis algorithm and a spatial clustering algorithm.

6. The real-scene three-dimensional based video twin garden human-vehicle flow security analysis method according to claim 1, characterized in that: The method comprises the following steps: Step one: acquiring multiple video stream data; Step two: applying a geometric transformation algorithm to correct video distortion, and fusing the corrected video data to a three-dimensional digital twin scene based on the established mapping relationship; Step three: identifying personnel targets in a video based on a target detection algorithm, and applying a three-dimensional coordinate conversion model to output personnel three-dimensional space position coordinates; Step four: using a streaming calculation framework to update the dynamic association between physical personnel and virtual objects in a three-dimensional scene in real time.

7. The real-scene three-dimensional-based video twin park human-vehicle flow security analysis method according to claim 6, characterized in that: The three-dimensional coordinate conversion model applies an intrinsic matrix, an extrinsic matrix and ground constraints of a camera for coordinate conversion.

8. The real-scene three-dimensional-based video twin park human-vehicle flow security analysis method according to claim 6, characterized in that: The dynamic association update includes applying a Kalman filtering algorithm and a Hungarian algorithm for multi-target tracking. 9.The real-scene three-dimensional based video twin garden human-vehicle flow security analysis method according to claim 6, characterized in that: The spatio-temporal correlation analysis includes extracting personnel moving speed parameters, direction parameters and stay time parameters.

10. The real-scene three-dimensional based video twin garden human-vehicle flow security analysis method according to claim 6, characterized in that: The spatio-temporal correlation analysis includes generating personnel flow trend prediction results based on historical trajectory data and real-time position data.

Images