Real-time video analysis method and system based on intelligent image processing

Abstract

The invention provides a real-time video analysis method and system based on intelligent image processing, and the method comprises the steps: obtaining the multi-view video data of a traffic monitoring region and the motion track data of a vehicle; performing inter-frame offset correction on the multi-view video data by using an optical flow compensation algorithm to generate corrected multi-view video data; extracting geometric feature points from the corrected multi-view video data, and constructing a topological relation graph based on the geometric feature points; the topological relation graph is optimized based on the movement track data, and an optimized topological relation graph is generated; and generating a video analysis result based on the optimized topological relation graph and the corrected multi-view video data. According to the invention, the accuracy and real-time performance of multi-view video analysis in a complex traffic scene are improved.

Description

Technical Field

[0001] The present application relates to the technical field of multi-view video real-time analysis and three-dimensional scene reconstruction in the field of intelligent traffic monitoring, and in particular to a real-time video analysis method and system based on intelligent image processing. Background Art

[0002] In the fields of intelligent urban traffic management and security monitoring, large-scale open scenes require real-time and accurate analysis of multi-view video data to obtain dynamic traffic flow information, detect abnormal events, and reconstruct 3D scenes. These scenarios require video analysis methods that can handle perspective differences between multiple cameras, compensate for moving objects, and ensure data consistency in complex environments, while also balancing real-time performance and robustness.

[0003] Currently, a typical solution uses multi-view video stitching technology based on feature point matching. This method extracts stable feature points from each camera's video, calculates a homography matrix for viewpoint alignment, and then uses a motion target detection algorithm to separate dynamic foreground objects from static backgrounds. Finally, the multi-channel video is fused to produce a panoramic surveillance image. This method relies on high-precision feature matching and uses offline calibration parameters to optimize the stitching effect.

[0004] This solution is prone to splicing misalignment in dynamic scenes due to the displacement of moving objects. Feature point matching is significantly affected by lighting changes and occlusions, and its reliance on offline calibration makes it difficult to adapt to real-time changes in camera pose. Furthermore, the incomplete separation of moving objects from the background can cause ghosting, affecting the accuracy of subsequent analysis. Summary of the Invention

[0005] The present application provides a real-time video analysis method and system based on intelligent image processing to solve the problems of low accuracy and poor real-time performance of multi-view video analysis in complex traffic scenes in the prior art.

[0006] In a first aspect, the present application provides a real-time video analysis method based on intelligent image processing, comprising: Acquire multi-view video data of the traffic monitoring area and vehicle motion trajectory data; Using an optical flow compensation algorithm, performing inter-frame offset correction on the multi-view video data to generate corrected multi-view video data; Extracting geometric feature points from the corrected multi-view video data, and constructing a topological relationship graph based on the geometric feature points; Optimizing the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph; A video analysis result is generated based on the optimized topological relationship graph and the corrected multi-view video data.

[0007] Optionally, the optimizing the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph includes: extracting a trajectory position sequence from the motion trajectory data; Mapping each position point in the trajectory position sequence to a corresponding node in the topological relationship graph; For any two adjacent nodes directly connected by a topological edge in the topological relationship graph, when the angle deviation between the moving direction angle values ​​of the two adjacent nodes and the preset direction angle value of the topological edge corresponding to the topological edge connecting the two adjacent nodes exceeds a preset angle threshold, the constraint weight value of the topological edge is updated; Based on the updated constraint weight values, the weight information of the topological edges in the topological relationship graph is optimized to generate an optimized topological relationship graph.

[0008] Optionally, when the angle deviation between the moving direction angle value of the two adjacent nodes and the preset direction angle value of the topological edge corresponding to the two adjacent nodes exceeds a preset angle threshold, updating the constraint weight value of the topological edge includes: Calculating a moving direction angle value of the two adjacent nodes, and using the moving direction angle value as a first direction value; The angle value of the preset direction of the topological edge connecting the two adjacent nodes is used as the second direction value; Calculating an absolute difference between the first direction value and the second direction value; If the absolute difference exceeds a preset angle threshold, the constraint weight value of the topological edge corresponding to the two adjacent nodes is multiplied by an incremental coefficient to generate an updated constraint weight value.

[0009] Optionally, generating a video analysis result based on the optimized topological relationship graph and the corrected multi-view video data includes: Position matching of the geometric feature points in the corrected multi-view video data with the nodes of the optimized topological relationship graph; If the matching fails, the geometric feature point corresponding to the matching failure is determined as a new geometric feature point, and the minimum Euclidean distance between the new geometric feature point and all existing nodes is calculated; When the minimum Euclidean distance exceeds a preset distance threshold, a new node corresponding to the newly added geometric feature point is inserted into the topological relationship graph, and a connection relationship set is updated; Based on the motion trajectory data, verify whether a connection path between the new node and all existing nodes is established; If the verification finds that there is at least one motion trajectory path that is inconsistent with the preset connected path in the updated topological relationship graph, it is determined that there is a conflict, and the constraint weight value of the relevant topological edge is adjusted to obtain an adjusted topological relationship graph; A video analysis result is generated according to the corrected multi-view video data, the adjusted topological relationship graph, and the motion trajectory data.

[0010] Optionally, inserting a new node corresponding to the newly added geometric feature point into the topological relationship graph and updating the connection relationship set includes: Obtaining a set of neighboring nodes within a preset radius of the new node; Creating a new node in the topological relationship graph, and setting the position coordinates of the new node to the coordinates of the newly added geometric feature point; A connection relationship is established between the new node and each node in the neighboring node set, and the connection relationship is added to the connection relationship set.

[0011] Optionally, constructing a topological relationship graph based on the geometric feature points includes: Based on the spatial positions of the geometric feature points, the geometric feature points corresponding to the same physical position are clustered into a position feature set; generating a connection relationship set according to the spatial adjacency relationship between adjacent position feature sets; A topological relationship graph is constructed with the location feature set as nodes and the connection relationship set as topological edges.

[0012] Optionally, performing inter-frame offset correction on the multi-view video data using an optical flow compensation algorithm to generate corrected multi-view video data includes: Calculating inter-frame motion vectors between adjacent video frames in the multi-view video data using an optical flow compensation algorithm; Filtering a valid motion region set from the inter-frame motion vectors; Recursively accumulating the offsets of the effective moving area set, and calculating an inter-frame offset correction parameter based on the accumulation result; Adjusting the position coordinates of the video frame according to the inter-frame offset correction parameter; The multi-view video data is corrected according to the adjusted video frame to generate corrected multi-view video data.

[0013] In a second aspect, the present application provides a real-time video analysis system based on intelligent image processing, comprising: An acquisition module is used to acquire multi-view video data of the traffic monitoring area and vehicle motion trajectory data; A first generating module is configured to perform inter-frame offset correction on the multi-view video data using an optical flow compensation algorithm to generate corrected multi-view video data; A construction module, configured to extract geometric feature points from the corrected multi-view video data and construct a topological relationship graph based on the geometric feature points; A second generating module is used to optimize the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph; The third generating module is used to generate a video analysis result based on the optimized topological relationship graph and the corrected multi-view video data.

[0014] In a third aspect, the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a real-time video analysis method based on intelligent image processing as described in any one of the first aspects.

[0015] In a fourth aspect, the present application provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a real-time video analysis method based on intelligent image processing as described in any one of the first aspects.

[0016] In the present application, a real-time video analysis method based on intelligent image processing is provided, which includes: obtaining multi-perspective video data of a traffic monitoring area and motion trajectory data of a vehicle; using an optical flow compensation algorithm to perform inter-frame offset correction on the multi-perspective video data to generate corrected multi-perspective video data; extracting geometric feature points from the corrected multi-perspective video data, and constructing a topological relationship graph based on the geometric feature points; optimizing the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph; and generating a video analysis result based on the optimized topological relationship graph and the corrected multi-perspective video data.

[0017] The technical solution provided by this application has the following beneficial effects: This application enables the simultaneous acquisition of multi-source heterogeneous data, providing a complete data foundation for subsequent analysis. It effectively eliminates inter-frame video displacement errors caused by camera shake or rapid object movement. It accurately captures key static features in the scene, providing a reliable benchmark for spatial modeling. It establishes a quantifiable and analyzable spatial relationship model for the road network. It uses real-world motion data to calibrate theoretical models, improving the accuracy of spatial representation. It outputs comprehensive monitoring data that integrates spatial topology and real-time dynamic information.

[0018] Furthermore, the present application also extracts the trajectory position sequence from the motion trajectory data and maps it to the topological relationship graph nodes. For any two adjacent nodes in the graph directly connected by a topological edge, when the deviation between their moving direction angle value and the preset direction angle value of the corresponding topological edge exceeds a threshold, the constraint weight value of the topological edge is dynamically updated, and finally a new topological relationship graph is generated based on the updated weight information optimization.

[0019] In addition, this application realizes dynamic graph optimization based on real motion trajectories. Through direction consistency verification and weight adaptive adjustment mechanism, it effectively corrects the deviation between the theoretical topological model and the actual traffic flow, and improves the accuracy of road network space representation and the ability to identify abnormal driving behavior.

[0020] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 A flowchart of a real-time video analysis method based on intelligent image processing provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a real-time video analysis system based on intelligent image processing provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0024] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0025] Existing multi-view video stitching technology based on feature point matching has significant shortcomings: in dynamic scenes, the displacement of moving objects can easily lead to stitching misalignment, feature point matching is significantly affected by lighting changes and occlusion, and reliance on offline calibration makes it difficult to adapt to real-time changes in camera pose. Furthermore, incomplete separation of moving objects from the background can cause ghosting, affecting the accuracy of subsequent analysis. These shortcomings stem from the fact that existing solutions lack sufficient compensation for dynamic offsets between video frames and a lack of topological optimization mechanisms for integrating motion trajectory data, resulting in reduced reliability of video analysis results in complex scenes.

[0026] To address the above-mentioned issues, this application proposes a real-time video analysis method based on intelligent image processing. By acquiring multi-view video data of a traffic monitoring area and vehicle motion trajectory data, the optical flow compensation algorithm is used to correct the inter-frame offset of the multi-view video data to generate corrected multi-view video data; geometric feature points are extracted from the corrected video data and a topological relationship graph is constructed; the topological relationship graph is optimized based on the motion trajectory data to generate an optimized topological relationship graph; and finally, the optimized topological relationship graph and the corrected multi-view video data are combined to generate video analysis results. This method effectively eliminates inter-frame offsets in dynamic scenes through optical flow compensation and optimizes the topological relationship graph using motion trajectory data. This solves problems such as splicing misalignment, illumination interference, and offline calibration dependence in the existing technology, thereby improving the accuracy and real-time performance of video analysis in complex scenes.

[0027] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0028] Figure 1 A flowchart of a real-time video analysis method based on intelligent image processing is provided in an embodiment of the present application. Figure 1As shown, the method includes: Step 101: Acquire multi-view video data of a traffic monitoring area and vehicle motion trajectory data.

[0029] In step 101, multi-view video data refers to video sequences captured synchronously from multiple cameras within a traffic monitoring area, containing road scene information from different angles. The video frames from each camera are strictly synchronized in time. Motion trajectory data represents continuous vehicle position information acquired by onboard positioning devices or roadside sensors, including attributes such as timestamps, longitude and latitude coordinates, and direction of movement.

[0030] In an embodiment of the present application, the system first deploys multiple wide-angle cameras to cover the target monitoring area, ensuring that the fields of view of each camera have sufficient overlapping areas; at the same time, it accesses the real-time trajectory data transmitted by the vehicle-mounted positioning device, establishes a unified timestamp to achieve spatiotemporal alignment of video data and trajectory data; the video acquisition module synchronously captures the images of each camera at a fixed frame rate, and the trajectory receiving module parses the vehicle position information in real time, and finally outputs time-aligned multi-source data for subsequent processing.

[0031] For example, at the intersection of a main road in a certain city, four high-definition cameras are installed to cover the four directions of east, west, south and north respectively. The cameras synchronously capture video at 25 frames per second. At the same time, the roadside unit receives the on-board GPS data of passing vehicles. The data contains vehicle ID, latitude and longitude coordinates and speed information. The system controls the time deviation between video frames and trajectory data to milliseconds through the network time protocol, ensuring the spatiotemporal consistency of subsequent processing.

[0032] Step 102: Using an optical flow compensation algorithm, perform inter-frame offset correction on the multi-view video data to generate corrected multi-view video data.

[0033] In step 102, the optical flow compensation algorithm analyzes the motion vectors of pixels in adjacent video frames to create an inter-frame displacement model. Inter-frame offset correction eliminates positional deviations between consecutive video frames caused by camera shake or rapid object movement. The corrected multi-view video data refers to a video sequence that has undergone inter-frame offset correction.

[0034] In an embodiment of the present application, for each camera's video stream, two consecutive frames of images are selected to extract feature points; the displacement vectors of the feature points in the previous and next frames are calculated to form an optical flow field; the main movement direction of the optical flow vector is counted, and abnormal offsets are filtered out; an affine transformation model is established based on the effective optical flow vector, and subsequent frame images are geometrically transformed according to the model parameters, and a position-corrected video sequence is output.

[0035] For example, in the video captured by the east-facing camera, the system detected slight camera shake between frames 100 and 101 due to strong winds. Using an optical flow algorithm, the system calculated the displacement of 10 stable feature points. After eliminating two outliers, the system took the average offset of 5 pixels horizontally and 3 pixels vertically for the eight valid points. This was then integrated with the historical data from the previous five frames (with a forgetting factor of α = 0.6) to obtain smoothing parameters Δx' = 4.8 pixels and Δy' = 2.9 pixels. A translation transformation model was then established to perform reverse compensation for the subsequent 102 frames and beyond, eliminating the effects of shake. This compensated video maintains the stability of stationary objects.

[0036] Step 103: extracting geometric feature points from the corrected multi-view video data, and constructing a topological relationship graph based on the geometric feature points.

[0037] In step 103, geometric feature points represent points in the video image that have significant geometric characteristics, such as building edge feature points, lane intersections, and traffic sign corners. A topological relationship graph is a graph structure composed of feature points as nodes and spatial connections between feature points as edges.

[0038] In an embodiment of the present application, stable geometric feature points are detected from the corrected video frames; feature points from multiple perspectives are matched and fused to eliminate duplicate points; points with close distances are clustered into key nodes based on the actual spatial distribution of feature points; and directed edges are established between adjacent nodes based on the actual road connections, with the direction of the edge representing the legal driving direction of the vehicle, ultimately generating a topological representation of the road network.

[0039] For example, the system extracts 32 lane intersections and 16 traffic sign corners from videos taken from four different viewpoints. After coordinate conversion and deduplication, 20 unique spatial feature points are retained. Based on the actual road layout, these points are connected to form a topological graph with 15 directed edges. The east-west main roads are bidirectional, while the side roads are one-way. The system also includes a mechanism for expanding the topological graph to accommodate temporary road changes, such as construction.

[0040] Step 104: Optimizing the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph.

[0041] In step 104, the optimized topological relationship diagram refers to the road network model after correction of the motion trajectory data, in which the nodes represent traffic key points, the edges represent the traffic paths, and the constraint weight values ​​of the edges have been adjusted according to the actual traffic flow direction deviation, which can more accurately reflect the actual road usage.

[0042] In an embodiment of the present application, motion trajectory data is mapped to the nodes of the topological graph; the direction of trajectory motion passing through each edge is counted; the average deviation between the trajectory direction and the preset direction of the edge is calculated; when the deviation exceeds a threshold, the constraint weight of the edge is increased proportionally; after traversing all edges and completing the weight adjustment, unreliable edges with too low weights are deleted, and the optimized topological graph is output.

[0043] For example, the system analyzed 1,000 trajectory data points and found that 12% of the preset one-way edges of a certain branch road had reverse driving trajectories, with an average direction deviation of 25 degrees; the constraint weight of this edge was increased from 1.0 to 1.8; and a temporary channel edge with only three trajectories passing through it was deleted. The final optimized topology map is more in line with actual traffic conditions.

[0044] Step 105: Generate a video analysis result based on the optimized topological relationship graph and the corrected multi-view video data.

[0045] In step 105 , the video analysis result represents dynamic analysis data of the fused topology structure and real-time video, including information such as traffic flow statistics and abnormal event detection.

[0046] In an embodiment of the present application, moving targets in real-time video are associated with topological map nodes; the legality of driving is determined based on the target's movement path in the topological map; traffic flow change trends are predicted in combination with historical trajectory data; areas with abnormal changes in topological edge weights are detected as potential event points; and a comprehensive analysis report containing violation warnings and congestion predictions is output.

[0047] For example, if the system detects that a truck has continuously violated the directional constraints of two one-way edges, it will immediately trigger a reverse traffic alarm. At the same time, based on the trend of continuously decreasing edge weights on the west-to-east main road, it predicts that congestion will occur in that direction and generates an early warning message that is pushed to the traffic management center.

[0048] This method achieves accurate video analysis in complex traffic scenarios through multi-source data fusion and dynamic optimization. Optical flow compensation ensures the stability of video data, a topological graph provides a structured road representation, and motion trajectory optimization enables the model to continuously adapt to actual traffic flow changes. The resulting system can reliably identify traffic violations and accurately predict traffic flow trends, improving the real-time and accuracy of intelligent traffic management.

[0049] In order to solve the problem of dynamic optimization of the topology relationship graph in complex traffic scenarios, in some embodiments, step 104: optimizing the topology relationship graph based on the motion trajectory data to generate the optimized topology relationship graph includes: Step 201: extracting a trajectory position sequence from the motion trajectory data.

[0050] In step 201 , the trajectory position sequence refers to a set of position data of a vehicle at consecutive time points, including timestamps, coordinate information, and movement directions, and is used to reflect the actual driving path of the vehicle.

[0051] In the embodiment of the present application, the system extracts continuous and valid position points from the original trajectory data, filters out stationary points and abnormal jump points, generates a smooth trajectory sequence in chronological order, and provides normalized input data for subsequent mapping.

[0052] Step 202: Map each position point in the trajectory position sequence to a corresponding node in the topological relationship graph.

[0053] In step 202, a location point refers to a single coordinate point in the vehicle's trajectory. It is derived from continuous location data collected by the vehicle's GPS device and includes information such as timestamps, longitude, and latitude. The corresponding node is the key road point in the topological relationship graph that is closest to the trajectory location point in spatial distance. The nearest neighbor is determined by calculating the Euclidean distance between the location point and all nodes. The node mapping process matches the trajectory location point to the nearest node in the topological relationship graph. Nodes in the topological relationship graph represent key road points, such as intersection centers or lane endpoints.

[0054] In an embodiment of the present application, the spatial distance between each trajectory position point and all nodes is calculated, and the nearest node is found as the mapping target. When multiple consecutive position points are mapped to the same node, deduplication processing is performed to ensure that each trajectory forms a coherent path representation in the graph.

[0055] Step 203: For any two adjacent nodes in the topological relationship graph that are directly connected by a topological edge, when the angle deviation between the moving direction angle value of the two adjacent nodes and the preset direction angle value of the topological edge corresponding to the topological edge connecting the two adjacent nodes exceeds a preset angle threshold, the constraint weight value of the topological edge is updated.

[0056] In step 203, adjacent nodes refer to two nodes in the topological relationship graph that are directly connected by an edge. Whether they are adjacent is determined based on the road connection relationship set during the initial mapping. The moving direction angle value refers to the actual driving direction angle of the vehicle when passing through two adjacent nodes, and is calculated clockwise with the north direction as the reference. The angle deviation refers to the angle difference between the actual movement direction of the vehicle and the preset direction of the topological edge. The preset angle threshold represents the critical value for determining whether the trajectory direction is in violation of the rules, and is pre-set according to the road grade and traffic management requirements. The constraint weight value represents the numerical value of the topological edge direction constraint strength. The initial value is 1.0, and it is dynamically adjusted according to the trajectory direction deviation.

[0057] In an embodiment of the present application, all trajectory directions passing through each topological edge are counted, and their average deviation from the preset direction of the edge is calculated. When the deviation continues to exceed a threshold, the constraint weight value of the edge is proportionally increased to enhance the ability to identify abnormal driving behavior.

[0058] Step 204: Based on the updated constraint weight values, the weight information of the topological edges in the topological relationship graph is optimized to generate an optimized topological relationship graph.

[0059] In step 204, the optimization process is the process of adjusting the topological graph connection relationship according to the updated constraint weights. The weight information is the set of constraint weight values ​​of all edges in the topological relationship, which is used to represent the strictness of the traffic rules of each path.

[0060] In the embodiment of the present application, the system traverses all topological edges, removes unreliable connections with weights below a critical value, merges redundant edges with similar directions, and finally generates a concise and reliable optimized topological relationship graph.

[0061] Here's a specific example: During the optimization process at a major urban arterial intersection, the system first extracted a continuous trajectory position sequence from received vehicle GPS data. It then filtered out static points with dwell times exceeding a threshold and abnormal coordinate changes to obtain a valid vehicle path. The cleaned trajectory positions were then mapped to nodes in a constructed topological graph. The preset direction angles for the east-west main road's bidirectional edges were 90 and 270 degrees, respectively, while the preset direction angle for the northbound branch's unidirectional edges was 180 degrees. Counting 1,000 valid trajectory data revealed that 120 trajectories on the northbound branch exhibited reverse driving behavior. The calculated average direction angle of these trajectories was 0 degrees, a deviation of 180 degrees from the preset direction angle of 180 degrees. Based on the preset angle threshold of 30 degrees, this deviation significantly exceeded the threshold. The system then adjusted the constraint weight for the edge using the formula W_new = W_old × 1.5, where W_old is the original weight of 1.0 and W_new is updated to 1.5. At the same time, it was detected that a temporary edge connecting a branch road and a parking lot was only used by three trajectories. The usage frequency was less than the set ratio threshold of the total number of trajectories. The system removed this edge from the topology map. After optimization, the new topology map accurately reflects the actual rules of two-way traffic on the main road and one-way traffic on the branch road, and strengthens the constraints on high-frequency violation sections.

[0062] In an embodiment of the present application, a dynamic optimization mechanism driven by trajectory data is used to enable the topological relationship diagram to continuously adapt to actual traffic flow changes, effectively improving the reliability of path compliance analysis in complex scenarios while maintaining the ability to respond quickly to changes in the road network.

[0063] In order to solve the dynamic optimization problem of directional constraints in a topological relationship graph, in some embodiments, step 203: when the angle deviation between the moving direction angle value of the two adjacent nodes and the preset direction angle value of the topological edge connecting the two adjacent nodes exceeds a preset angle threshold, updating the constraint weight value of the topological edge includes: Step 301: Calculate the moving direction angle value of the two adjacent nodes, and use the moving direction angle value as the first direction value.

[0064] In step 301, the moving direction angle value is calculated by calculating the direction vector from the trajectory coordinates between adjacent nodes and taking the inverse tangent to obtain the angle. For example, if a vehicle passes through nodes A (100, 50) and B (105, 55), the direction vector is (5, 5), and arctan (5 / 5) = 45 degrees. The first direction value is the actual moving direction angle of the vehicle when passing between two adjacent nodes, calculated from the trajectory coordinates.

[0065] In the embodiment of the present application, the system extracts continuous position points passing through the two adjacent nodes from the trajectory position sequence, calculates the vehicle movement direction based on the change in the coordinates of the position points, and converts it into a standard angle representation.

[0066] Step 302: taking the angle value of the preset direction of the topological edge corresponding to the two adjacent nodes as the second direction value.

[0067] In step 302, the preset direction angle refers to the legal travel direction angle of the road pre-set in the topological relationship diagram, reflecting the vehicle's travel direction required by traffic regulations. The second direction value refers to the legal travel direction angle pre-set for the topological edge, which is set according to traffic regulations during map construction.

[0068] In the embodiment of the present application, the system reads the direction parameter stored in the topological edge attribute, which is set according to the actual road traffic rules during initial mapping.

[0069] Step 303: Calculate the absolute difference between the first direction value and the second direction value.

[0070] In step 303 , the absolute difference refers to the minimum angular difference between the first direction value and the second direction value, and is used to quantify the degree of deviation between the actual driving direction and the specified direction.

[0071] In the embodiment of the present application, the system calculates the difference between the two angle values ​​and takes the absolute value. If the result is greater than 180 degrees, the value is subtracted from 360 degrees to ensure that the minimum angle difference is obtained.

[0072] Step 304: If the absolute difference exceeds a preset angle threshold, the constraint weight value of the topological edge corresponding to the two adjacent nodes is multiplied by an incremental coefficient to generate an updated constraint weight value.

[0073] In step 304, the incremental coefficient is used to dynamically adjust the constraint weight value. Different coefficients are set based on the road importance level. The updated constraint weight value represents the adjusted constraint strength of the edge direction. A larger value indicates stricter monitoring of the edge direction.

[0074] In the embodiment of the present application, the system maintains a coefficient comparison table. When it is detected that the angle deviation exceeds the limit, the corresponding coefficient is selected according to the road grade for weight adjustment.

[0075] Here's a specific example: In a traffic monitoring system at a major urban arterial intersection, the system dynamically optimizes the one-way traffic rules for a northbound branch road. First, 120 reverse-travel trajectories are obtained, and the angular values ​​of each trajectory as it passes through adjacent nodes are calculated. The average value is then taken to obtain a first angular value of 0 degrees, derived from the inverse tangent of the trajectory's start and end coordinates. Simultaneously, the preset angular value of 180 degrees for the branch road edge in the topological relationship is read as the second angular value. The system calculates the absolute difference between the two values ​​to obtain 180 degrees. Normalization is performed using the formula Δθ = min|θ1-θ2|, 360 - |θ1-θ2|, ultimately determining the actual angular deviation to be 180 degrees. Because this value exceeds the preset angular threshold of 30 degrees, the system selects an incremental coefficient of 1.5 based on the branch level. The constraint weight is updated using the formula W_new = W_old × k, where W_old is the original weight of 1.0 and k is the incremental coefficient of 1.5, resulting in a new weight of 1.5.

[0076] In the embodiment of the present application, through the dynamic direction comparison and weight adjustment mechanism, the topological relationship diagram can adapt to the changes in traffic flow. While maintaining the rigidity of road rules, it implements key monitoring of high-frequency violation sections, effectively improving the intelligent level of traffic management.

[0077] In order to improve the dynamic adaptability and analysis accuracy of the traffic monitoring system, in some embodiments, step 105: generating video analysis results based on the optimized topological relationship graph and the corrected multi-view video data includes: Step 401: Positionally match the geometric feature points in the corrected multi-view video data with the nodes in the optimized topological relationship graph.

[0078] In step 401, the position matching process refers to the process of comparing the geometric feature points detected in the video data with the spatial positions of existing nodes in the topological relationship graph.

[0079] In the embodiment of the present application, the system calculates the spatial distance between the coordinates of the video feature points and the coordinates of the topological nodes. When the distance is less than the matching threshold, it is determined that the match is successful, otherwise it enters the new node processing flow.

[0080] Step 402: If the matching fails, the geometric feature point corresponding to the failed matching is determined as a new geometric feature point, and the minimum Euclidean distance between the new geometric feature point and all existing nodes is calculated.

[0081] In step 402, newly added geometric feature points are road feature points that appear in the video but are not registered in the topology map. Existing nodes are road key points that already exist in the topology map and are derived from historical feature point clustering results. The minimum Euclidean distance is the shortest straight-line distance between the point and all nodes in the topology map.

[0082] In the embodiment of the present application, the system traverses and calculates the square root of the sum of the squares of the coordinate differences between the new feature point and each existing node, and records the minimum value as the basis for judgment.

[0083] Step 403: When the minimum Euclidean distance exceeds a preset distance threshold, a new node corresponding to the newly added geometric feature point is inserted into the topological relationship graph, and the connection relationship set is updated.

[0084] In step 403, the preset distance threshold (e.g., 5 meters) is used to determine whether to add a new node. This threshold is set based on road density and monitoring accuracy requirements. New nodes represent newly discovered road feature points in the video and are added to the topology map after distance verification. The process of adding a new node involves assigning a unique ID to the new feature point, setting its coordinate attributes, and searching for neighboring nodes to establish connections.

[0085] In this embodiment, if the minimum Euclidean distance does not exceed the threshold, the newly added feature point is classified into the feature set of the nearest node, and no new node is created. The system searches for existing nodes within the radius of the new node, establishes bidirectional or unidirectional connecting edges based on the actual road connection, and initializes the weight values.

[0086] Step 404: Based on the motion trajectory data, verify whether the connection paths between the new node and all existing nodes are established.

[0087] In step 404, the connectivity path refers to a valid connection sequence between nodes in the topology graph. The connectivity path verification process refers to the process of verifying whether the motion trajectory supports the newly established topological connection relationship.

[0088] In the embodiment of the present application, the system retrieves historical trajectory data, searches for trajectory records that pass through both the new node and the adjacent node, and calculates the degree of matching between the movement direction and the topological edge direction.

[0089] Step 405: If verification finds that there is at least one motion trajectory path that is inconsistent with the preset connected path in the updated topological relationship graph, it is determined that there is a conflict, and the constraint weight value of the relevant topological edge is adjusted to obtain the adjusted topological relationship graph.

[0090] In step 405, the inconsistency of the connected path means that the edge that the trajectory actually passes through does not exist in the graph, or the deviation of the trajectory direction from the preset direction of the edge exceeds a threshold. "Inconsistency" refers to the following two situations: the first is that the edge that the trajectory actually passes through does not exist in the topological graph, such as the vehicle goes directly from node A to node C, but there is only ABC path in the graph; the second is that the angular deviation between the trajectory direction and the preset direction of the edge exceeds a threshold, such as the edge is preset to be one-way eastward, and the trajectory is westward. Connected path conflict judgment means that there is a non-negligible deviation between the actual movement direction of the trajectory and the preset direction of the topological edge. Related topological edges refer to edges that the trajectory actually passes through but do not exist in the graph, as well as edges between new nodes and adjacent nodes. The adjusted topological relationship graph refers to the road network model after verification of the newly added nodes and correction of conflicting edges.

[0091] In the embodiment of the present application, the system calculates the ratio and deviation angle of the conflicting trajectory, adjusts the constraint weights of the relevant edges according to preset rules, and may delete the connecting edge in case of serious conflict.

[0092] Step 406: Generate a video analysis result based on the corrected multi-view video data, the adjusted topological relationship graph, and the motion trajectory data.

[0093] In an embodiment of the present application, the system integrates information such as violation identification, congestion prediction, and road network change detection into a structured report.

[0094] Here's a specific example: During road construction at a main road intersection in a certain city, the system discovered a new corner feature point of a temporary construction fence in the east-facing camera's field of view while processing the corrected multi-view video data in real time. The coordinates of this feature point were matched with 20 existing nodes in the optimized topological relationship diagram, and the Euclidean distance between this feature point and the nearest node, the east-bound crosswalk sign, was calculated as d = sqrt[(x2-x1)²+(y2-y1)²] = 6.3 meters, where x1 and y1 are the coordinates of the crosswalk node, exemplified by (x1, y1) = (0, 0), and x2 and y2 are the coordinates of the new feature point, exemplified by (x2, y2.) = (6.3, 0). Because this distance exceeded the preset 5-meter threshold, the system determined it to be a newly added geometric feature point and created a corresponding new node in the topological diagram. With this node as the center, a search was conducted for a set of neighboring nodes within a 7-meter radius, including the east-bound crosswalk node and the southeast corner streetlight node. Based on actual road conditions, the system established a bidirectional edge connecting the new node to the crosswalk node, with an initial weight of 1.0. It also established a unidirectional edge connecting the new node to the streetlight node. The system then retrospectively analyzed the most recent 1,000 trajectory data and found that 35 trajectories supported connectivity between the new node and the crosswalk node, with consistent directions. However, only two trajectories supported connectivity with the streetlight node, with mixed directions. Based on the verification results, the system deleted the edge connecting to the streetlight node and increased the weight of the edge connecting to the crosswalk node to 1.2, resulting in an adjusted topology. Based on this adjusted topology, the system detected in real time a construction vehicle illegally entering the crosswalk area from the new node (with a directional deviation of 65 degrees, exceeding the 30-degree threshold). An alert for the construction vehicle violation was immediately generated. Simultaneously, analysis of the trajectory data stream revealed a 20% decrease in traffic speed on the eastbound main road, predicting congestion within 10 minutes. The system then automatically sent an alert containing the specific location coordinates and a recommended detour route to the traffic management center. The final output of the video analysis results includes: first, real-time violation event reports (vehicle ID, violation type, timestamp); second, dynamic road condition predictions (congestion area, expected duration); third, topology map change records (newly added node coordinates, connection edge adjustment details).

[0095] In the embodiment of the present application, by dynamically adding and deleting nodes and verifying connection relationships, the system can adapt to changes in the road network, ensuring that the traffic analysis results are always consistent with the actual situation, while improving the accuracy of identifying violations and abnormal conditions.

[0096] In order to improve the dynamic update mechanism of the topology relationship graph in the traffic monitoring system, in some embodiments, step 403: inserting a new node corresponding to the newly added geometric feature point in the topology relationship graph and updating the connection relationship set includes: Step 501: Obtain a set of neighboring nodes within a preset radius of the new node.

[0097] In step 501, the neighboring node set refers to a set of all road key point nodes existing in the topological relationship graph within a circular area with the new node as the center and a preset radius as the range.

[0098] In an embodiment of the present application, the system quickly retrieves existing nodes that meet the distance conditions through spatial indexing. These nodes represent characteristic points of traffic elements such as intersections and lane line endpoints.

[0099] Step 502: Create a new node in the topological relationship graph, and set the position coordinates of the new node to the coordinates of the newly added geometric feature point.

[0100] In step 502 , the position coordinates of the new node refer to the spatial position data of the newly added geometric feature point in the video coordinate system.

[0101] In the embodiment of the present application, the system maps the coordinates of the feature points detected by the video to the topological relationship through pre-calibrated conversion parameters. Figure 1 The global coordinate system ensures the consistency of spatial position.

[0102] Step 503: Establish a connection relationship between the new node and each node in the neighboring node set, and add the connection relationship to the connection relationship set.

[0103] In step 503, establishing a connection relationship involves determining the direction and type (unidirectional / bidirectional) of the connection between the new node and its neighboring nodes based on actual road traffic rules, and initializing the weight value. Adding a connection relationship involves adding the newly created edge object (including direction and weight attributes) to the edge collection data structure of the topology graph.

[0104] In an embodiment of the present application, the system establishes connection edges with directionality and weight attributes between the new node and each adjacent node based on the road type and traffic control information, and initializes the constraint weight values ​​of these edges.

[0105] Here's a specific example: During road construction at a major urban intersection, the system detected a new corner feature point of a temporary construction fence using an east-facing camera. After conversion, its coordinates were determined to be x = 125.4 meters and y = 78.6 meters. The system then searched for neighboring nodes within a preset 7-meter radius, centered around this point. The calculated coordinates for the east-bound crosswalk node were x = 126.0 meters and y = 78.2 meters, with a distance of d = sqrt[(126.0-125.4)² + (78.2-78.6)²] = 0.72 meters. The southeast corner streetlight node had coordinates of x = 126.2 meters and y = 78.8 meters, with a distance of d = sqrt[(126.2-125.4)² + (78.8-78.6)²] = 0.82 meters, both within the search range. After creating a new node and setting its coordinates, the system, based on the construction plan design, determines that the temporary fence and the crosswalk should maintain two-way traffic. Therefore, a bidirectional connection is established with an initial weight of 1.0. Due to the physical separation from the streetlight node, only a one-way connection is established for monitoring purposes. The newly added connection is recorded in the connection relationship set, with the preset directions for the bidirectional connection set to 120 degrees from the fence to the crosswalk and 300 degrees in the opposite direction, respectively, and the one-way connection set to 60 degrees from the fence to the streetlight.

[0106] In the embodiment of the present application, through standardized new node insertion and connection relationship establishment processes, it is ensured that the topological relationship diagram can respond to changes in road layout in a timely manner, maintain the traffic monitoring system's ability to accurately model the dynamic environment, and provide reliable support for real-time traffic analysis and decision-making.

[0107] In order to establish a topological model that accurately reflects the road network structure, in some embodiments, step 103: constructing a topological relationship graph based on the geometric feature points includes: Step 601: Based on the spatial positions of the geometric feature points, the geometric feature points corresponding to the same physical position are clustered into a position feature set.

[0108] In step 601, spatial location refers to the coordinates of a geometric feature point in a unified coordinate system, obtained by converting camera calibration parameters. The same physical location refers to multiple video feature points corresponding to the same real-world traffic element (e.g., intersection center). A location feature set is the clustering result of geometric feature points belonging to the same traffic element in spatial location.

[0109] In the embodiment of the present application, the system uses a spatial density clustering algorithm to merge geometric feature points such as lane line intersections, traffic sign corners, etc. that are close in distance and have similar features into a feature set representing the same intersection or road key points.

[0110] Step 602: Generate a connection relationship set based on the spatial neighbor relationship between adjacent position feature sets.

[0111] In step 602, the connection relationship set refers to the road connectivity relationship between different location feature sets, and the sum of the connection relationships between all nodes in the topological graph, including attributes such as edge direction and weight.

[0112] In the embodiment of the present application, the system analyzes the spatial distribution and road directions between each feature set, establishes a connection relationship between the location feature sets with actual traffic possibility, and records the connection direction and initial weight.

[0113] Step 603: construct a topological relationship graph with the location feature set as nodes and the connection relationship set as topological edges.

[0114] In an embodiment of the present application, the system uses the clustered feature set as a graph node, establishes weighted directed edges based on the connection relationship set, and ultimately forms a topological model that fully describes the road network.

[0115] Here's a specific example: During the implementation of an intelligent transportation system at a major urban road intersection, the system processed stable video data captured by four cameras. Eight lane intersections were first extracted from the east camera, with coordinates ranging from (x1, y1) to (x8, y8). Seven were extracted from the west camera, six from the south camera, and five from the north camera. After a coordinate system transformation, a density clustering algorithm was used to group feature points with spatial distances less than 2 meters into positional feature sets. This resulted in four main sets: the east entrance set contained three feature points, with the center coordinates calculated by arithmetic mean (125.6, 78.3); the west entrance set contained two feature points, with the center coordinates (75.4, 78.5); the south entrance set contained three feature points, with the center coordinates (100.2, 55.1); and the north entrance set contained two feature points, with the center coordinates (100.5, 101.7). Based on the actual road design, the system creates bidirectional edges between east-west intersections, with angles of 90 and 270 degrees, respectively; bidirectional edges between south-north intersections, with angles of 0 and 180 degrees; and one-way edges between each entrance lane and its corresponding turn lane set, such as a 210-degree angle between the east entrance lane and the south turn lane. Each edge has an initial weight of 1.0, with an additional weight of 0.2 for turn lanes. The resulting topological diagram contains four nodes and eight edges, accurately reflecting the actual traffic rules at the intersection.

[0116] In the embodiment of the present application, the automatic conversion from video data to road network model is achieved through intelligent clustering of geometric feature points and construction of topological relationships, providing a structured road network foundation for subsequent traffic analysis and anomaly detection.

[0117] To improve the stability of the video data, in some embodiments, step 102: using an optical flow compensation algorithm to perform inter-frame offset correction on the multi-view video data to generate corrected multi-view video data, includes: Step 701: Calculate inter-frame motion vectors between adjacent video frames in the multi-view video data using an optical flow compensation algorithm.

[0118] In step 701, adjacent video frames refer to two frames with consecutive time stamps, and the adjacent relationship is determined by the frame number sequence. The inter-frame motion vector refers to the displacement direction and size of the corresponding feature point in adjacent video frames.

[0119] In an embodiment of the present application, the system analyzes the position changes of feature points of fixed objects in two consecutive frames of images, calculates the movement of each feature point in the horizontal and vertical directions, and forms a vector field that describes the overall inter-frame motion.

[0120] Step 702: Filter a valid motion region set from the inter-frame motion vectors.

[0121] In step 702, the effective motion region screening process involves compiling the directional histograms of all motion vectors, retaining regions whose directions fall within ±15 degrees of the main peak, and excluding isolated, abnormal regions. The effective motion region set is defined as continuous image regions with consistent motion directions. Validity is determined by directional consistency and spatial continuity.

[0122] In the embodiment of the present application, the system counts the directional distribution of all inter-frame motion vectors, marks the areas where the direction deviates from the mainstream direction by less than a threshold as valid, and excludes abnormal motion areas caused by local occlusion or noise.

[0123] Step 703: recursively accumulate the offsets of the effective motion region set, and calculate an inter-frame offset correction parameter based on the accumulation result.

[0124] In step 703, the offset refers to the positional difference between video frames and is calculated using the effective moving area vector. The recursive accumulation process is the weighted fusion of the effective offset of the current frame and the historical offsets. The inter-frame offset correction parameter is the displacement that needs to be compensated for the video frame.

[0125] In the embodiment of the present application, the system adopts a sliding window mechanism to perform weighted averaging on the effective offsets of the most recent frames, smooth out instantaneous fluctuations, and obtain a stable inter-frame offset correction parameter.

[0126] Step 704: Adjust the position coordinates of the video frame according to the inter-frame offset correction parameter.

[0127] In step 704, the position coordinates refer to the positions of the video pixels in the image coordinate system, which are obtained by feature point detection. The process of adjusting the position coordinates refers to the process of performing geometric transformation on the video frame according to the correction parameters.

[0128] In the embodiment of the present application, the system establishes an affine transformation model, converts the correction parameters into a transformation matrix, and performs reverse displacement compensation on the video frame.

[0129] Step 705: Correct the multi-view video data according to the adjusted video frame to generate corrected multi-view video data.

[0130] In step 705 , the video data correction process refers to applying the calculated geometric transformation parameters (such as translation matrix) to each frame of the image and resampling to generate a stable video sequence.

[0131] In the embodiment of the present application, the system applies correction parameters to all video frames in sequence to ensure that the static background remains stable throughout the entire video sequence.

[0132] Here's a specific example: In a surveillance scenario at a major urban intersection, the east-facing camera experienced periodic jitter due to strong winds. When processing video data from frames 100 to 101, the system selected 10 stable feature points, such as traffic sign corners and streetlight bases. Using an optical flow algorithm, the system calculated the motion vectors for each point. Eight of these feature points showed downward and rightward movement, with an average lateral offset of Δx = 5 pixels and a vertical offset of Δy = 3 pixels. The calculation formulas were Δx = 1 / n∑(xi_101 - xi_100) and Δy = 1 / n∑(yi_101 - yi_100), where n = 8 is the number of valid feature points and xi and yi are the feature point coordinates. After eliminating two anomalous feature points, the system then weightedly fused the offsets of the eight valid motion regions with the historical data from the previous five frames. Using a forgetting factor of α = 0.6, the system recursively calculated the smoothed correction parameters, Δx' = 4.8 pixels and Δy' = 2.9 pixels. A translation transformation matrix M = [10-Δx'; 01-Δy'] is established to perform geometric correction on the 102nd frame and subsequent video frames.

[0133] In the embodiment of the present application, through the optical flow compensation and dynamic correction mechanism, the video instability problem caused by external interference such as camera jitter is effectively eliminated, ensuring the accuracy of subsequent topological relationship construction and traffic analysis, and improving the reliability of the system in complex environments.

[0134] Figure 2 A schematic diagram of the structure of a real-time video analysis system based on intelligent image processing provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes: The acquisition module 21 is used to acquire multi-view video data of the traffic monitoring area and the motion trajectory data of the vehicle.

[0135] The first generating module 22 is configured to perform inter-frame offset correction on the multi-view video data by using an optical flow compensation algorithm to generate corrected multi-view video data.

[0136] The construction module 23 is configured to extract geometric feature points from the corrected multi-view video data and construct a topological relationship graph based on the geometric feature points.

[0137] The second generating module 24 is configured to optimize the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph.

[0138] The third generating module 25 is configured to generate a video analysis result based on the optimized topological relationship graph and the corrected multi-view video data.

[0139] Figure 2 The real-time video analysis system based on intelligent image processing can perform Figure 1 The implementation principles and technical effects of the real-time video analysis method based on intelligent image processing described in the illustrated embodiment are not further elaborated. The specific manner in which the various modules and units perform operations in the real-time video analysis system based on intelligent image processing in the aforementioned embodiment have been described in detail in the relevant embodiments of the method and will not be further elaborated here.

[0140] In one possible design, Figure 2 A real-time video analysis system based on intelligent image processing in the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0141] The processing component 32 is used to perform the above Figure 1 The embodiment provides a real-time video analysis method based on intelligent image processing.

[0142] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0143] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0144] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0145] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0146] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0147] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0148] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The embodiment shown is a real-time video analysis method based on intelligent image processing.

[0149] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0150] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0151] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time video analysis method based on intelligent image processing, characterized in that: include: Acquire multi-view video data of the traffic monitoring area and vehicle motion trajectory data; Using an optical flow compensation algorithm, performing inter-frame offset correction on the multi-view video data to generate corrected multi-view video data; Extracting geometric feature points from the corrected multi-view video data, and constructing a topological relationship graph based on the geometric feature points; Optimizing the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph; A video analysis result is generated based on the optimized topological relationship graph and the corrected multi-view video data.

2. The method according to claim 1, characterized in that The optimizing the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph includes: extracting a trajectory position sequence from the motion trajectory data; Mapping each position point in the trajectory position sequence to a corresponding node in the topological relationship graph; For any two adjacent nodes directly connected by a topological edge in the topological relationship graph, when the angle deviation between the moving direction angle values ​​of the two adjacent nodes and the preset direction angle value of the topological edge corresponding to the topological edge connecting the two adjacent nodes exceeds a preset angle threshold, the constraint weight value of the topological edge is updated; Based on the updated constraint weight values, the weight information of the topological edges in the topological relationship graph is optimized to generate an optimized topological relationship graph.

3. The method according to claim 2, characterized in that When the angle deviation between the moving direction angle value of the two adjacent nodes and the preset direction angle value of the topological edge corresponding to the two adjacent nodes exceeds a preset angle threshold, updating the constraint weight value of the topological edge includes: Calculating a moving direction angle value of the two adjacent nodes, and using the moving direction angle value as a first direction value; The angle value of the preset direction of the topological edge connecting the two adjacent nodes is used as the second direction value; Calculating an absolute difference between the first direction value and the second direction value; If the absolute difference exceeds a preset angle threshold, the constraint weight value of the topological edge corresponding to the two adjacent nodes is multiplied by an incremental coefficient to generate an updated constraint weight value.

4. The method according to claim 1, wherein Generating a video analysis result based on the optimized topological relationship graph and the corrected multi-view video data includes: Position matching of the geometric feature points in the corrected multi-view video data with the nodes of the optimized topological relationship graph; If the matching fails, the geometric feature point corresponding to the matching failure is determined as a new geometric feature point, and the minimum Euclidean distance between the new geometric feature point and all existing nodes is calculated; When the minimum Euclidean distance exceeds a preset distance threshold, a new node corresponding to the newly added geometric feature point is inserted into the topological relationship graph, and a connection relationship set is updated; Based on the motion trajectory data, verify whether a connection path between the new node and all existing nodes is established; If the verification finds that there is at least one motion trajectory path that is inconsistent with the preset connected path in the updated topological relationship graph, it is determined that there is a conflict, and the constraint weight value of the relevant topological edge is adjusted to obtain an adjusted topological relationship graph; A video analysis result is generated according to the corrected multi-view video data, the adjusted topological relationship graph, and the motion trajectory data.

5. The method according to claim 4, characterized in that The inserting a new node corresponding to the newly added geometric feature point into the topological relationship graph and updating the connection relationship set includes: Obtaining a set of neighboring nodes within a preset radius of the new node; Creating a new node in the topological relationship graph, and setting the position coordinates of the new node to the coordinates of the newly added geometric feature point; A connection relationship is established between the new node and each node in the neighboring node set, and the connection relationship is added to the connection relationship set.

6. The method according to claim 1, characterized in that The constructing of a topological relationship graph based on the geometric feature points includes: Based on the spatial positions of the geometric feature points, the geometric feature points corresponding to the same physical position are clustered into a position feature set; generating a connection relationship set according to the spatial adjacency relationship between adjacent position feature sets; A topological relationship graph is constructed with the location feature set as nodes and the connection relationship set as topological edges.

7. The method according to claim 1, characterized in that The method of using an optical flow compensation algorithm to perform inter-frame offset correction on the multi-view video data to generate corrected multi-view video data includes: Calculating inter-frame motion vectors between adjacent video frames in the multi-view video data using an optical flow compensation algorithm; Filtering a valid motion region set from the inter-frame motion vectors; Recursively accumulating the offsets of the effective moving area set, and calculating an inter-frame offset correction parameter based on the accumulation result; Adjusting the position coordinates of the video frame according to the inter-frame offset correction parameter; The multi-view video data is corrected according to the adjusted video frame to generate corrected multi-view video data.

8. A real-time video analysis system based on intelligent image processing, characterized in that: include: An acquisition module is used to acquire multi-view video data of the traffic monitoring area and vehicle motion trajectory data; A first generating module is configured to perform inter-frame offset correction on the multi-view video data using an optical flow compensation algorithm to generate corrected multi-view video data; A construction module, configured to extract geometric feature points from the corrected multi-view video data and construct a topological relationship graph based on the geometric feature points; A second generating module is used to optimize the topological relationship graph based on the motion trajectory data to generate an optimized topological relationship graph; The third generating module is used to generate a video analysis result based on the optimized topological relationship graph and the corrected multi-view video data.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a real-time video analysis method based on intelligent image processing as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the real-time video analysis method based on intelligent image processing according to any one of claims 1 to 7 is implemented.

Images