Intersection motor vehicle conflict identification method, electronic equipment and storage medium

Abstract

The invention relates to a conflict identification method for motor vehicles at an intersection, electronic equipment and a storage medium, which are applied to a road intersection monitoring system and used for acquiring initial video data acquired by N monitoring cameras at the same time, restoring all pixel coordinates of a frame image in the initial video data into real coordinates through a perspective matrix, and identifying the conflict of the motor vehicles at the intersection. Obtaining a calibration image; carrying out motor vehicle identification in a preset range of the calibration image to obtain motor vehicle data of N shooting angles, and placing the motor vehicle data of different shooting angles in corresponding matching pools; matching the motor vehicle data in the N matching pools according to a preset sequence, searching motor vehicles appearing in the plurality of monitoring cameras, and storing the corresponding motor vehicle data in a result pool; according to the motor vehicle data in the result pool, track intersection points between the vehicles are analyzed, and according to the time and speed change characteristics of the motor vehicles passing through the track intersection points, whether the motor vehicles have the conflict risk or not is judged.

Description

Technical Field

[0001] The present application relates to the technical field of traffic safety monitoring, and in particular to a method, electronic device, and storage medium for identifying conflicts between motor vehicles at intersections. Background Art

[0002] Intersection conflicts primarily occur when two or more traffic participants collide without slowing down or changing direction. Intersection conflicts are a key observation area in traffic safety research, helping to assess intersection safety, identify potential traffic safety hazards, and uncover the causes of traffic accidents. The study of intersection conflicts is of great significance in the field of traffic safety. Through systematic and detailed conflict analysis, it not only improves intersection safety management but also provides a scientific basis for the early identification and prevention of traffic safety hazards, thereby effectively reducing the occurrence of traffic accidents and ensuring road traffic safety.

[0003] Currently, conflict identification at intersections is mainly performed through analysis of images acquired by a single camera. However, the viewing angle of a single camera is limited because it cannot cover the entire area of ​​an intersection or road section, resulting in some traffic participants (such as obscured vehicles and pedestrians) not being detected. If a wide-angle lens is used, the wide-angle lens will introduce perspective distortion, affecting the accurate calculation of the target position and speed. Therefore, existing conflict detection methods are limited by equipment and algorithm limitations, and the data accuracy and reliability are often low, making it difficult to fully and accurately reflect the actual situation of traffic conflicts.

[0004] In existing traffic monitoring systems, multiple surveillance cameras are installed at intersections to cover the traffic conditions in the intersection area. The application of intersection monitoring mainly focuses on traffic management and public security prevention, such as real-time monitoring of traffic flow, capturing traffic violations, intelligent management of traffic signals, and criminal behavior monitoring and evidence collection. However, there is currently no solution for intersection conflict identification based on multiple monitoring perspectives. Summary of the Invention

[0005] The embodiments of the present application provide a method, electronic device, and storage medium for identifying conflicts between motor vehicles at an intersection, so as to at least solve the problem in the related art of lacking intersection conflict identification based on multiple monitoring perspectives.

[0006] In a first aspect, an embodiment of the present application provides a method for identifying conflicts between motor vehicles at an intersection, which is applied to a road intersection monitoring system, wherein the monitoring system includes N monitoring cameras, each of which has a different shooting angle, where N is an integer not less than 2. The method includes: Obtaining initial video data from N surveillance cameras, and restoring pixel coordinates of frame images in the initial video data to real-world coordinates to obtain a calibration image; Performing motor vehicle identification within a preset range of the calibration image to obtain motor vehicle data and cache it in a preset matching pool, wherein data from different surveillance cameras are placed in different matching pools; The motor vehicle data in each matching pool is matched cyclically in a preset order to search for motor vehicles that appear in multiple surveillance cameras, and the corresponding motor vehicle data is saved in the result pool; The trajectory intersections between the vehicles are analyzed based on the motor vehicle data in the result pool, and whether the motor vehicles have a collision risk is determined based on the time and speed change characteristics of the motor vehicles passing through the trajectory intersections.

[0007] In one embodiment, restoring pixel coordinates of the frame image in the initial video data to real-world coordinates to obtain a calibrated image includes: Obtaining pixel coordinate values ​​and real-world coordinate values ​​of a plurality of reference points in the frame image, wherein the plurality of reference points are not collinear; The pixel coordinate values ​​and the real-world coordinate values ​​are combined into a matrix equation, and the least squares method is used to solve the equation to obtain a perspective transformation matrix; The real-world coordinates of other pixel points in the frame image are calculated according to the perspective transformation matrix, and the frame image is converted from the pixel coordinate system to the real-world coordinate system to obtain multiple frames of calibration images.

[0008] In one embodiment, the performing of vehicle identification within a preset range of the calibration image, obtaining vehicle data, and caching the data in a preset matching pool includes: Obtaining a pre-built optimized YOLO algorithm model, performing motor vehicle recognition on the registered calibration image based on the optimized YOLO algorithm model, identifying the motor vehicle and assigning a temporary ID; Tracking the identified motor vehicles using the IoU indicator to obtain trajectory information of all motor vehicles in the calibration image; The temporary ID and trajectory information of the motor vehicle are combined into motor vehicle data and cached in a corresponding matching pool.

[0009] In one embodiment, the construction of the optimized YOLO algorithm model includes: Obtain a basic YOLO algorithm architecture, where the basic YOLO algorithm architecture includes a multi-scale feature fusion layer and a post-processing optimization module; A quarter-scale layer is added to the multi-scale feature fusion layer. The detailed features of the quarter-scale layer are fused through upsampling and the CSP double convolution bottleneck module. A space-to-depth conversion module is introduced to compress the spatial information into the depth channel, expand the receptive field, and retain the edge features of small objects. Embedding a selection kernel attention module in the multi-scale feature fusion layer, and adding a coordinate attention module after the quarter-scale layer; In the post-processing optimization module, motion perception suppression is performed on motor vehicle identification through an adaptive non-maximum suppression mode, and noise filtering is performed through a multimodal trajectory filtering method.

[0010] In one embodiment, the step of cyclically matching the vehicle data in each matching pool in a preset order, searching for vehicles that appear in multiple surveillance cameras, and saving the corresponding vehicle data to a result pool includes: Selecting a matching pool as a reference matching pool, cyclically matching the motor vehicle data in the reference matching pool with the motor vehicle data in other matching pools to determine whether the same motor vehicle is captured and identified by more than two surveillance cameras; If so, the motor vehicle data corresponding to the motor vehicle is placed in the result pool; if not, the motor vehicle data in the candidate pool is matched again. If the match is successful, the motor vehicle data corresponding to the motor vehicle is placed in the result pool; if the match is unsuccessful, the motor vehicle data is placed in the candidate pool; Each matching pool is configured with a corresponding candidate pool, and the candidate pool is used to store motor vehicle data that has not been successfully matched.

[0011] In one embodiment, determining whether the same motor vehicle is photographed and identified by more than two surveillance cameras includes: During the matching process, a piece of motor vehicle data is extracted from each of the two matching pools, wherein the motor vehicle data includes a center coordinate value and a movement direction of the motor vehicle; Calculating the Euclidean distance according to the center coordinate values, and calculating the deflection angles of the two motor vehicles according to the movement directions; When the Euclidean distance and the deflection angle meet preset requirements, the two motor vehicles are determined to be the same physical entity, and a unique global ID is assigned to the physical entity.

[0012] In one embodiment, the method further comprises: After the global ID is assigned, the motor vehicle data is saved to the result pool using a queue and hash table data structure; The vehicle data stored in the candidate pool is used to match the vehicle data in the next matching calculation cycle. If the match is successful, a global ID is assigned and placed in the result pool; If no match is successful within the preset time window, the invalid motor vehicle data will be cleared.

[0013] In one embodiment, the vehicle data includes a timestamp and speed change characteristics of the frame image. The determining whether the motor vehicle has a collision risk based on the time and speed change characteristics of the motor vehicle passing through the trajectory intersection includes: Obtain any trajectory intersection, filter all vehicles in the result pool that pass through the trajectory intersection, and record the global ID of the vehicle, the timestamp when it passes through the trajectory intersection, and the speed change characteristics; The time difference between multiple motor vehicles when passing through the trajectory intersection is calculated based on the timestamp. If the time difference is less than a preset time threshold, it is determined based on the speed change characteristics whether the acceleration of at least one motor vehicle is negative within a preset time range. If so, it is determined that there is a risk of collision between the motor vehicles at the trajectory intersection.

[0014] In a second aspect, an embodiment of the present application provides a computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for identifying conflicts among motor vehicles at intersections as described in the first aspect above is implemented.

[0015] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying conflicts between motor vehicles at an intersection as described in the first aspect above.

[0016] The method for identifying collisions between motor vehicles at an intersection, the electronic device, and the storage medium provided by the embodiments of the present application have at least the following technical effects: The intersection motor vehicle conflict identification method of the present application can help to quickly and accurately identify conflicts at filmed intersections at low cost and obtain the basic data required for research. The present application obtains the perspective transformation matrix based on the intersection monitoring video and reference points to restore the real coordinate information of the intersection. Then, the frame-by-frame trajectory of the vehicle at the intersection is obtained through the YOLO algorithm and the IoU threshold, and multi-perspective cross-camera target matching and position reconstruction are achieved. The coordinates of possible conflict points are obtained based on the intersection of the trajectories, and the vehicles passing the target points are sorted. The time threshold and speed change characteristics are combined to realize the identification of the conflicting subjects, and finally complete the identification of traffic conflicts at urban intersections under the monitoring perspective. It can not only quickly and accurately identify intersection conflicts at low cost, but also provide high-precision data to help the development of traffic management and safety research.

[0017] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 This is a flow chart of a method for identifying conflicts between motor vehicles at an intersection according to an embodiment of the present application; Figure 2 is a schematic diagram of reference points of a frame image captured by any surveillance camera in an embodiment of the present application; Figure 3 This is a schematic diagram of the monitoring perspective of the intersection monitoring system in one embodiment of the present application; Figure 4 This is a flow chart of matching motor vehicle data in N matching pools in one embodiment of the present application; Figure 5 yes Figure 4 Flowchart of the subsequent matching process for matching failed data in the embodiment; Figure 6 This is a schematic diagram of trajectory set screening in one embodiment of the present application; Figure 7 This is a schematic diagram of conflict set screening in one embodiment of the present application; Figure 8 It is a structural block diagram of an electronic device in one embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0020] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0021] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0022] Unless otherwise defined, technical or scientific terms used herein shall have the ordinary meaning as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "an," "the," and similar expressions used herein do not denote quantitative limitations and may refer to either the singular or the plural. The terms "comprise," "include," "have," and any variations thereof, used herein, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or modules (units) is not limited to the listed steps or units but may also include steps or units not listed, or may include other steps or units inherent to the process, method, product, or apparatus. The terms "connected," "connected," "coupled," and similar expressions used herein are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used herein, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" may mean: A exists alone; A and B exist simultaneously; or B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0023] Currently, intersection monitoring images cover the entire intersection, but there are no applications for conflict identification, nor are there any low-cost solutions for utilizing this data source. The current manual conflict identification process primarily involves subjectively identifying conflicts, extracting opposing trajectories, and outputting the coordinates of the conflict point and the speed of the conflicting parties, or the time, coordinates, and speed of trajectory changes. This presents challenges such as subjective conflict assessment, a cumbersome process, inaccurate identification, and poorly accurate output data.

[0024] Based on the intersection monitoring perspective, the present invention provides a low-cost data extraction method for traffic safety analysis in intersection scenarios. It calculates the perspective transformation matrix to restore the intersection coordinates, and realizes multi-perspective cross-camera target matching and position reconstruction. Through accurate traffic conflict recognition and frame-by-frame data accuracy, it realizes intersection conflict assessment, helps identify intersection safety hazards, improves the level of intersection traffic safety management, improves the design of intersection safety facilities, and optimizes signal phase settings.

[0025] Based on the above situation, an embodiment of the present application provides a method, electronic device and storage medium for identifying conflicts between motor vehicles at an intersection.

[0026] In a first aspect, an embodiment of the present application provides a method for identifying conflicts between motor vehicles at an intersection, which is applied to a road intersection monitoring system, wherein the monitoring system includes N monitoring cameras, and different monitoring cameras have different shooting angles, wherein N is an integer not less than 2, and reference Figure 1 , the conflict identification method of this application is implemented by the following steps.

[0027] Step S1: Obtain initial video data from N surveillance cameras, and restore pixel coordinates of frame images in the initial video data to real-world coordinates to obtain a calibrated image.

[0028] In one embodiment, it is necessary to first obtain pixel coordinate values ​​and real-world coordinate values ​​of multiple reference points in the frame image, wherein the multiple reference points are not collinear. Figure 2 It is a schematic diagram of the reference point of one of the shooting angles in this embodiment; then the pixel coordinate values ​​and the real-world coordinate values ​​are combined into a matrix equation, which is solved and calculated by the least squares method to obtain a perspective transformation matrix; finally, the real-world coordinates of other pixel points in the frame image are calculated according to the perspective transformation matrix, and the frame image is converted from the pixel coordinate system to the real-world coordinate system to obtain a multi-frame calibration image.

[0029] Specific reference Figure 2 , the pixel coordinates of 5 points in the known image (coordinates in the image coordinate system) and their corresponding real-world coordinates (coordinates in the target coordinate system).

[0030] Let the perspective transformation matrix be M, which is a 3×3 matrix:

[0031] For a point (x, y) in an image, the homogeneous coordinates are represented as [x, y, 1] T , the corresponding point in the target coordinate system after perspective transformation is (X, Y), and the homogeneous coordinates are expressed as [X, Y, 1] T , the relationship between them satisfies:

[0032] In the embodiment of the present application, the five reference points in the figure are combined into a matrix equation and solved using the least squares method to obtain the perspective transformation matrix M.

[0033] For each pixel (x I ,y I ), we can get the corresponding coordinates (X′, Y′) of the image I′ in the real coordinate system, satisfying the relationship:

[0034] in,

[0035] Then calculate the coordinates in the real world (X I , Y I ):

[0036] In the actual calculation process, the video stream contains a large number of continuous frames per second. If multi-camera target recognition and matching are performed on each frame, it will greatly consume computing resources and pose challenges to real-time and system scalability. To this end, this application adopts frame extraction processing technology, which aims to select key frames per second for image processing and conflict identification. For example, only 1-2 frames are processed per second, and the remaining frames are skipped and only used for local caching or simple tracking.

[0037] The advantages of this method are reflected in the following aspects: improving computing efficiency: greatly reducing the number of frames to be processed, reducing CPU / GPU pressure, and ensuring that the system can run efficiently on ordinary servers and even devices; ensuring data representativeness: using equally spaced frame extraction can ensure that each vehicle is captured at least once when passing through the monitoring area, achieving full coverage and reducing the risk of missed detections; simplifying data deduplication: vehicles appear multiple times in multiple frames, and frame extraction reduces the complexity of ID duplication technology, laying a good foundation for subsequent trajectory organization and statistical analysis.

[0038] Step S2: Perform motor vehicle recognition within a preset range of the calibration image, obtain motor vehicle data, and cache the data in a preset matching pool, wherein data from different surveillance cameras are placed in different matching pools.

[0039] Specifically, a pre-built optimized YOLO algorithm model is obtained, and motor vehicle recognition is performed on the aligned calibration image based on the optimized YOLO algorithm model, and motor vehicles are identified and assigned temporary IDs; the identified motor vehicles are tracked using the Intersection over Union (IoU) indicator to obtain trajectory information of all motor vehicles in the calibration image; and motor vehicle data is combined according to the temporary ID and trajectory information of the motor vehicles and cached in a corresponding matching pool.

[0040] Among them, the construction of the optimized YOLO algorithm model of this embodiment includes: obtaining a basic YOLO algorithm architecture, which includes a multi-scale feature fusion layer and a post-processing optimization module; adding a quarter-scale layer to the multi-scale feature fusion layer, fusing the detail features of the quarter-scale layer through upsampling and a CSP double convolution bottleneck module, i.e., a C2f module, and introducing a space-to-depth conversion module to compress spatial information into a depth channel, expand the receptive field and retain small target edge features; embedding a selection kernel attention module in the multi-scale feature fusion, and adding a coordinate attention module after the quarter-scale; in the post-processing optimization module, motion perception suppression is performed for motor vehicle recognition through an adaptive non-maximum suppression mode, and noise filtering is performed through a multimodal trajectory filtering method.

[0041] In this embodiment of the present application, YOLO is used to identify vehicles and IoU-assisted output of frame-by-frame vehicle trajectories. YOLO is a single-stage object detection algorithm whose core concept is to transform the object detection task into a single end-to-end regression problem, directly predicting bounding boxes and class probabilities on the image grid. Because the original YOLOv8 model is not effective for identifying vehicles at intersections under surveillance, improvements are now made to the model structure in terms of multi-scale feature enhancement and post-processing optimization.

[0042] On the one hand, this embodiment achieves multi-scale feature enhancement by adding a P2 layer deep fusion architecture. More specifically, this application adds a P2 layer (1 / 4 scale) based on the original P3-P5 layers. Through upsampling and the CSP double convolution bottleneck module, namely the C2f module, it fuses shallow detail features and introduces a spatial-to-depth (SPD) conversion module to compress spatial information into the depth channel, expand the receptive field, and preserve the edge features of small objects.

[0043] Original input feature map After space compression changes:

[0044] Output dimension , effectively avoiding the problem of small target feature loss in downsampling, and improving the recall rate of small target detection in traffic monitoring scenarios:

[0045] In addition, the embodiment of the present application also embeds a selection kernel (SK) attention module in PAN-FPN, using 3×3 and 5×5 dual-branch convolution to capture multi-scale features; and strengthens key areas of the vehicle (such as lights, tires, etc.) through channel weight self-adaptation.

[0046] In order to enhance spatial attention, a coordinate attention (CA) module can be added after the P2 layer to improve direction perception.

[0047] Assume input feature F in , dual-branch convolution output:

[0048] Channel weight calculation:

[0049] Final feature fusion:

[0050] Where σ is the sigmoid function. This design improves the model's ability to extract features of key vehicle components from a monitoring perspective.

[0051] In terms of the post-processing optimization system, the embodiment of the present application adopts adaptive non-maximum suppression and multimodal trajectory filtering.

[0052] The specific suppression scheme is as follows: 1. Category-aware Soft-NMS, which performs suppression independently for vehicle categories (cars, trucks, etc.); 2. Use Gaussian decay function ; 3. Motion state perception suppression: relax the IoU threshold for detection boxes with consistent motion trajectories in consecutive frames.

[0053] Specifically, the traditional NMS elimination strategy is shown in the following formula:

[0054] The improved motion perception suppression in the embodiment of the present application is as follows:

[0055] The speed consistency factor is:

[0056] The multimodal trajectory filtering method is shown in Table 1.

[0057] Table 1:

[0058] Vehicle state vector:

[0059] State transition equation:

[0060] Compared with the traditional uniform velocity model (4-latitude state), the increase in acceleration latitude reduces the trajectory prediction error:

[0061] Continuous-time conflict function:

[0062] in:

[0063] The parameters satisfy the constraints:

[0064] Issue a warning when the function value exceeds a threshold:

[0065] The multi-factor integration model reduces the false alarm rate by 57.3% compared with single-frame detection.

[0066] Step S3, cyclically matching the motor vehicle data in each matching pool in a preset order, searching for motor vehicles that appear in multiple surveillance cameras, and saving the corresponding motor vehicle data to a result pool.

[0067] Specifically, a matching pool is selected as a reference matching pool, and the motor vehicle data in the reference matching pool is cyclically matched with the motor vehicle data in other matching pools to determine whether the same motor vehicle is photographed and identified by more than two surveillance cameras; if so, the motor vehicle data corresponding to the motor vehicle is placed in a result pool; if not, the data is matched again with the motor vehicle data in the candidate pool, and if the match is successful, the motor vehicle data corresponding to the motor vehicle is placed in the result pool; if the match is unsuccessful, the motor vehicle data is placed in a candidate pool; wherein each matching pool is configured with a corresponding candidate pool, and the candidate pool is used to store motor vehicle data that has not been successfully matched.

[0068] The order of the loop can be set manually, refer to Figure 3 and Figure 4For example, four surveillance cameras correspond to four matching pools, namely matching pool G1, matching pool G2, matching pool G3, and matching pool G4, and the corresponding candidate pools are candidate pool GA1, candidate pool GA2, candidate pool GA3, and candidate pool GA4. Taking G1 as the base matching pool, each matching pool contains multiple motor vehicle data. The motor vehicle data in G1 is traversed and first matched with each data in G2 in a loop. If there is a successful match, the successful motor vehicle data is matched with the data in G3. If the match is successful in G3, it is matched with the data in G4. If the match is successful in G4, it is entered into the result pool GF. In addition, after the data in each matching pool is traversed, the data that did not successfully match in each matching pool is placed in the corresponding candidate pool. The above is a matching order provided in an embodiment of the present application. Users can also set matching rules according to accuracy requirements or time costs. For example, each matching pool is matched in pairs. If G1 and G2 have successfully matched data, they can be directly placed in the result pool without matching with G3 and G4; or, if G1, G2 and G3 have successfully matched data, they can also be directly placed in the matching pool without matching with G4. Therefore, the specific matching order and matching method are not limited here.

[0069] refer to Figure 5 , let the current cycle be i+1 and the previous cycle be i. If the match fails in multiple matching pools, the data C that failed to match will be i+1 First compare with the data C of the previous cycle in the candidate pool i Perform matching one by one, for example, the data that failed to match is matched with the motor vehicle data C in the candidate pool GA1, candidate pool GA2, candidate pool GA3, and candidate pool GA4 in sequence. i If the match is successful, the data corresponding to the temporary id (i.e., tracking id) is added to the result pool GF, and the data corresponding to the id is deleted from the candidate pool; if the match fails, the data corresponding to the tracking id is added to the corresponding candidate pool to form the motor vehicle data C of the current cycle i+1 .

[0070] In one embodiment, the vehicle data includes the timestamp of the frame image, the center coordinates of the vehicle in the image, the vehicle's temporary ID, direction of movement, license plate number, and confidence level. To determine whether other matching pools have vehicle data for the vehicle at the current moment, a piece of vehicle data is extracted from each of the two matching pools during the matching process. The vehicle data includes the center coordinates and direction of movement of the vehicle. A Euclidean distance is calculated based on the center coordinates, and a deflection angle between the two vehicles is calculated based on the direction of movement. When the Euclidean distance and deflection angle meet preset requirements, the two vehicles are determined to be the same physical entity and a unique global ID is assigned to the physical entity. The real geographic coordinates of each vehicle added to the result pool should be determined for each frame (by averaging the real geographic coordinates of the vehicle across all matching pools). Subsequent trajectory intersection analysis is based on this unique set of real geographic coordinate data.

[0071] Specifically, the matching search process of this embodiment is to perform multi-view cross-camera target matching and position reconstruction, wherein this embodiment calculates the number of view angles based on the actual camera layout at the intersection. Figure 3 Taking a conventional intersection as an example, four surveillance cameras were installed, capturing images from four different viewpoints. Four matching pools were designed based on the number of cameras, each corresponding to the object recognition results from the current frame. To associate the identity of the same vehicle from different viewpoints, the project employed spatial inversion and a multi-pool spatial matching mechanism.

[0072] The detailed implementation process is as follows. This embodiment uses the first matching pool as a benchmark, iterates through each vehicle candidate within it, and sequentially searches for its nearest neighbor in the other three matching pools within the global physical coordinate space. Specifically, a threshold interval based on Euclidean distance is used to determine whether two detection boxes belong to the same physical entity. Dynamic constraints such as motion direction are also incorporated to further improve matching accuracy. Only when multiple cameras find objects with highly consistent spatial positions are they considered to be the same vehicle and a unique global ID is assigned to each vehicle.

[0073] For the vehicles in the trajectory sets {G1, G2, G3, G4} in the four matching pools, they are recorded as {id_1_1, id_2_1, ....} (the vehicle id is named as id_number_matching pool number). Taking the first matching pool as the benchmark, traverse the vehicles in the trajectory set and obtain the real-world coordinates (x1, y1, z1) (the coordinates are the center coordinates of the vehicle) and the movement direction V_dir1 of vehicle id_number_1. Traverse the trajectory set G2 in the second matching pool, traverse the vehicles, obtain the registered coordinates (x2, y2, z2) and movement direction V_dir2 of id_number_2, match the coordinate positions of the vehicles in the two matching pools, and consider the length size of the real car and the recognition accuracy error. The vehicle matching coordinate error is set within 1 meter. Assuming that the registration coordinate distance is d1 and the real distance is d2, the conversion ratio is (unit m) is defined as:

[0074] The formula for target matching is:

[0075] This application fully utilizes the information redundancy in the spatial overlap of multiple cameras, not only improving the robustness of cross-viewpoint target matching, but also effectively resolving misidentification issues caused by single-viewpoint occlusion, illumination changes, and other issues. Ultimately, each vehicle receives a globally unique ID through this spatial matching process.

[0076] In a preferred embodiment, after the global ID is assigned, the motor vehicle data is saved to the result pool using a queue and hash table data structure; the motor vehicle data stored in the candidate pool is used to match the motor vehicle data at the next moment. If the match is successful, a global ID is assigned and placed in the result pool; if no match is successful within the preset time window, the invalid motor vehicle data is cleared.

[0077] Specifically, for targets that fail to complete multi-camera spatial matching (e.g., a vehicle captured by only two cameras or temporarily lost due to occlusion), the system temporarily stores these targets and their detection information in a "candidate pool." Subsequent, newly extracted frames will attempt to associate them with unmatched targets in the candidate pool. Once a new detection frame with a high degree of consistency in spatial location and dynamic features is found, the previously pending candidate targets are completed and incorporated into the global ID system.

[0078] This application considers the recognition of a keyframe as a calculation cycle Ci (or a preset duration of one calculation cycle, such as 1 second). When entering the next keyframe calculation cycle Ci+1, the trajectory IDs {GA1, GA2, GA3, GA4} in the candidate pool from the previous calculation cycle Ci will be retained. If a vehicle ID in Ci+1 is not matched and enters the candidate pool before entering the candidate pool, this ID will undergo an additional match with the candidate pool within the Ci cycle. If a match is successful, this ID will also be added to the result pool GF. If a match cannot be found within the preset time window, it is assumed that the target has left the monitoring scene or is a false detection. Invalid data can be regularly cleared to ensure a streamlined and efficient database.

[0079] In addition, under normal circumstances, a match is considered successful only if all four matching pools G1-G4 are successfully matched. Starting from G1, matching must continue until G4 before it is added to the result pool. However, considering camera angle issues or occlusion issues caused by tall vehicles, it is possible to adjust the matching process to only match 2 or 3 matching pools before adding them to the result pool. The addition method is also very simple: simply merge the candidate pool GA3 or GA4 into the result pool GF.

[0080] The matching search method in this embodiment ensures that all valid vehicle targets receive consistent, unique, and accurate identification. Each global ID not only associates spatial coordinates, appearance time, and other attribute information across frames, but also supports the formation of a complete behavioral trajectory record. This structured data not only serves real-time traffic management but also provides a solid data foundation for subsequent business applications such as big data analysis, abnormal behavior detection, and road utilization assessment.

[0081] Step S4: analyzing the trajectory intersections between vehicles based on the motor vehicle data in the result pool, and judging whether there is a collision risk between the motor vehicles based on the time and speed change characteristics of the motor vehicles passing through the trajectory intersections.

[0082] Specifically, before outputting the trajectory intersections, the trajectories of all motor vehicles in the video are drawn to obtain an initial trajectory set. For example, the initial trajectory set is C = {c1, c2, c3, c4, c5, c6, c7, c8}, and the existing intersection points p are detected. The detected intersection points are output as an intersection set P = {p1, p2, ..., p m}, including serial number and intersection coordinate information. Figure 6 After determining the intersection point, filter the passing motor vehicles and use the intersection point p j For example, 1≤j≤, filter all the points passing through the intersection p j The motor vehicle, according to the point p j Sort the vehicle trajectories in chronological order, record the vehicle ID information, passing point p j The timestamp information of the intersection is output as the trajectory set:

[0083] refer to Figure 6 , passing through the intersection point p j Traj_p j The method is described as follows: {c2, c4, c5, c7, c8}, and then, based on the time and speed change characteristics of the vehicles passing through the trajectory intersection, it is determined whether the vehicles are at risk of collision. Specifically, any trajectory intersection is obtained, and all vehicles passing through the trajectory intersection are screened from the result pool. The global ID of each vehicle, the timestamp when it passed through the trajectory intersection, and the speed change characteristics are recorded. The time difference between multiple vehicles passing through the trajectory intersection is calculated based on the timestamp. If the time difference is less than a preset time threshold, it is determined whether at least one of the vehicles has a negative acceleration within a preset time range. If so, it is determined that the vehicles are at risk of collision at the trajectory intersection.

[0084] Calculate the time difference of the timestamps of adjacent vehicle trajectories in the order of recording, and select the trajectories with a time difference less than or equal to the time threshold TA to form a trajectory pair (the time threshold TA is set according to actual needs); secondly, judge the speed change characteristics of the vehicles involved in the trajectory pair, and select the trajectory pairs where at least one of the two vehicles has a negative acceleration within the time threshold TA, that is, obtain the trajectory pairs at the intersection point p j Conflict set where traffic conflicts occur Information. Specific reference Figure 7 , calculate the time difference between adjacent trajectories, set the time threshold TA, and j The trajectories with Δt ≤ TA are selected to form trajectory pairs, such as (c2, c4), (c4, c5) and (c7, c8). Then, the speed characteristics are screened to ensure that at least one of the two vehicles has deceleration behavior within the time threshold.

[0085] Repeat the above process until all the intersections in the intersection set are conflict identified. The output information is the conflict identification within the intersection view, and the complete set of conflict pairs is output. .

[0086] In the preferred embodiment, long-term global vehicle ID merging is required because in real-world multi-camera traffic monitoring scenarios, vehicles enter and exit the scene dynamically and continuously. Some vehicles may appear and disappear in the frame, while others may experience transient ID fluctuations (i.e., the same vehicle may be assigned multiple IDs) due to occlusion, lighting, or detection errors. Therefore, designing an efficient and robust time-series ID merging and trajectory correction mechanism is crucial to the accuracy and usability of the system.

[0087] After each vehicle is assigned a globally unique ID, the system will use data structures such as queues and hash tables to orderly store and manage its spatial coordinates, timestamps, and matching status across frames. The specific approach is as follows: The queue stores the spatial coordinates and time information of vehicles passing through it, enabling sequential playback of their trajectories. The hash table, using vehicle IDs as keys, facilitates efficient access and retrieval of the status of all active vehicles within the current monitoring area. The sliding window mechanism, for large-scale, high-frequency data, periodically recycles data on departing vehicles through a sliding window, alleviating memory pressure. This structure not only supports fine-grained analysis of individual vehicle trajectories but also facilitates subsequent batch statistics and behavioral pattern mining.

[0088] In actual operation, affected by external factors, the system will inevitably encounter problems such as ID loss and misclassification. To this end, this system introduces two key technologies: trajectory continuity analysis and historical trajectory comparison, to perform anomaly detection and ID merging (anomaly recovery mechanism). First, trajectory continuity analysis is used to detect the correspondence between IDs between adjacent frames. If an ID is suddenly interrupted and a new ID appears near the scene, and its movement direction, speed, time interval, etc. are highly consistent with the original ID, it is inferred to be a short-term occlusion or ID drift, and the two trajectories are automatically merged. Second, historical trajectory comparison is used to further assist in merging misclassified IDs by comparing their historical spatial paths (such as trajectory overlap) and vehicle appearance features (such as color, size, etc.) for IDs that have repeated meanings over a long period of time.

[0089] Preferably, an exception completion strategy is also required to not immediately judge an ID that has not appeared for a short period of time as "leaving", but to set a buffer. During this period, if a vehicle that meets the trajectory pattern is redetected, the original ID can still be continued, thereby improving tracking continuity and fault tolerance.

[0090] This mechanism ensures that even in complex and dynamic environments, the same vehicle maintains a unique and complete trajectory over long time periods. Ultimately, the spatiotemporal trajectories and behavior records of all vehicles are uniformly recorded in the vehicle ID database, which can be used for a variety of subsequent applications, including traffic flow statistics, congestion analysis, and anomaly detection.

[0091] In a preferred embodiment, parameter tuning and adaptive mechanisms can be employed. This application flexibly adjusts trajectory merging criteria (e.g., maximum allowable time interval, spatial distance threshold, reducing the number of cameras, acquiring a smaller number of frames per second, etc.) based on road scene complexity, camera density, and computing resources to achieve an optimal balance between accuracy and efficiency. A manual intervention interface is also provided to facilitate manual correction of erroneous merging or deletion of IDs in special circumstances, further ensuring data quality.

[0092] This application can also flexibly adjust parameters such as the frame rate, pending pool duration (indicating the number of candidate pool matching calculation cycles (the number of C)), and spatial matching threshold according to actual needs to adapt to different road scenarios, traffic density, and hardware resource conditions, achieving the optimal balance of "high performance + high accuracy".

[0093] The spatial matching threshold refers to the allowable error in the coordinates when matching vehicles in different matching pools. Due to the camera angle, the real geographic coordinates of the same vehicle in different cameras may have slight differences. Assuming that the real geographic coordinates of vehicle 1 in camera No. 1 are (1, 1) and the real geographic coordinates in camera No. 2 are (2, 2), the Euclidean distance between the two coordinates is 1.4. Based on this assumption, if the spatial matching threshold is adjusted to 2, the two coordinates are determined to be the same vehicle. If the spatial matching threshold is set to 1, the two coordinates are determined to be different vehicles. By adjusting the spatial matching threshold, the accuracy of spatial vehicle matching can be determined.

[0094] This application provides a method for identifying motor vehicle conflicts at intersections, enabling collision recognition within existing monitoring systems. Using video footage captured by intersection surveillance cameras, this application uses advanced image processing algorithms to automatically identify and track the movement of motor vehicles within the surveillance video. Multi-view camera target matching and position reconstruction further improve accuracy, pinpointing potential conflict points.

[0095] Furthermore, this application utilizes deep learning and trajectory analysis technologies to accurately extract and calculate the various metrics required for conflict assessment, such as speed, acceleration, and path deviation. This meticulous analysis of the motion characteristics of each traffic participant ensures the high accuracy and reliability of the output data. These high-precision metrics not only provide a solid foundation for quantitative analysis of traffic conflicts but also provide critical data support for subsequent traffic safety research and improvement measures.

[0096] Furthermore, this application can quantitatively evaluate the effectiveness of traffic management measures or road design improvements by analyzing before-and-after traffic conflict data. Specifically, the system can collect and analyze data before and after the implementation of improvement measures. By comparing key indicators such as conflict frequency and severity, it can intuitively demonstrate the actual effects of the improvement measures. This function can provide traffic management departments with scientific and objective decision-making basis and promote the continuous improvement of traffic safety.

[0097] Finally, this application achieves low cost and high coverage by using fixed surveillance cameras (existing urban infrastructure). It can directly utilize the city's widely deployed surveillance equipment without the need for additional hardware investment, making it suitable for long-term large-scale applications. The use of a perspective transformation matrix (five reference points) has higher precision, solves the distortion problem of the camera's tilted perspective, and fits the real road space relationship. Multi-camera collaboration, cross-view target matching and position reconstruction, adapt to complex scenes, support intersections covered by multiple cameras, and solve occlusion and blind spot problems; comparison files rely on a single perspective and are easily affected by occlusion. The conflict judgment has a lower false alarm rate, and non-conflict scenarios (such as normal following) are filtered out in combination with behavioral characteristics (deceleration), making the judgment more consistent with the definition of traffic conflict.

[0098] Engineering feasibility: Balances computing efficiency and data integrity through frame extraction processing. High real-time performance: Reduces computing load through key frame processing, suitable for edge device deployment; the candidate pool mechanism improves the robustness of ID matching.

[0099] In summary, the proposed method for identifying motor vehicle conflicts at intersections not only enables motor vehicle conflict identification from an intersection monitoring perspective, provides low-cost, high-precision conflict assessment data, but also quantitatively evaluates the effects of pre- and post-processing improvements, significantly enhancing the scientific and practical nature of traffic conflict analysis. These technical results have broad application prospects in intelligent transportation systems and traffic safety research, providing strong support for future traffic management and safety strategy development.

[0100] In a second aspect, an embodiment of the present application provides an electronic device, Figure 8 FIG is a block diagram of an electronic device according to an exemplary embodiment. Figure 8 As shown, the electronic device may include a processor 11 and a memory 12 storing computer program instructions.

[0101] Specifically, the processor 11 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0102] The memory 12 may include a large-capacity memory for data or instructions. By way of example, and not limitation, the memory 12 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 12 may include removable or non-removable (or fixed) media. Where appropriate, the memory 12 may be internal or external to the data processing device. In certain embodiments, the memory 12 is non-volatile memory. In certain embodiments, the memory 12 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data out dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0103] The memory 12 may be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 11 .

[0104] The processor 11 reads and executes computer program instructions stored in the memory 12 to implement any one of the methods for identifying conflicts between motor vehicles at an intersection in the above embodiments.

[0105] In one embodiment, the electronic device may further include a communication interface 13 and a bus 10. Figure 8 As shown, the processor 11 , the memory 12 , and the communication interface 13 are connected via a bus 10 and communicate with each other.

[0106] The communication interface 13 is used to implement communication between the various modules, devices, units, and / or devices in the embodiments of the present application. The communication interface 13 can also implement data communication with other components such as: external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0107] The bus 10 includes hardware, software, or both, and couples the components of the electronic device to each other. The bus 10 includes, but is not limited to, at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 10 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Bus 10 may include one or more buses, where appropriate. Although embodiments herein describe and illustrate a particular bus, this application contemplates any suitable bus or interconnect.

[0108] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the method for identifying conflicts between motor vehicles at an intersection provided in the first aspect.

[0109] The readable storage medium may include, but is not limited to, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0110] In a possible embodiment, the present invention can also be implemented in the form of a program product, which includes program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps of the method for conflict identification of motor vehicles at an intersection provided in the first aspect.

[0111] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or entirely on the remote device.

[0112] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for identifying conflicts between motor vehicles at an intersection, characterized in that: The method is applied in a road intersection monitoring system, wherein the monitoring system includes N monitoring cameras, and different monitoring cameras have different shooting angles, wherein N is an integer not less than 2; the method includes: Obtaining initial video data from N surveillance cameras, and restoring pixel coordinates of frame images in the initial video data to real-world coordinates to obtain a calibration image; Performing motor vehicle identification within a preset range of the calibration image to obtain motor vehicle data and cache it in a preset matching pool, wherein data from different surveillance cameras are placed in different matching pools; The motor vehicle data in each matching pool is matched cyclically in a preset order to search for motor vehicles that appear in multiple surveillance cameras, and the corresponding motor vehicle data is saved in the result pool; The trajectory intersections between the vehicles are analyzed based on the motor vehicle data in the result pool, and whether the motor vehicles have a collision risk is determined based on the time and speed change characteristics of the motor vehicles passing through the trajectory intersections.

2. The method according to claim 1, characterized in that The step of restoring pixel coordinates of the frame image in the initial video data to real-world coordinates to obtain a calibration image includes: Obtaining pixel coordinate values ​​and real-world coordinate values ​​of a plurality of reference points in the frame image, wherein the plurality of reference points are not collinear; The pixel coordinate values ​​and the real-world coordinate values ​​are combined into a matrix equation, and the least squares method is used to solve the equation to obtain a perspective transformation matrix; The real-world coordinates of other pixel points in the frame image are calculated according to the perspective transformation matrix, and the frame image is converted from the pixel coordinate system to the real-world coordinate system to obtain multiple frames of calibration images.

3. The method according to claim 1, characterized in that The step of performing motor vehicle identification within a preset range of the calibration image, obtaining motor vehicle data and caching the data in a preset matching pool includes: Obtaining a pre-built optimized YOLO algorithm model, performing motor vehicle recognition on the registered calibration image based on the optimized YOLO algorithm model, identifying the motor vehicle and assigning a temporary ID; Tracking the identified motor vehicles using the IoU indicator to obtain trajectory information of all motor vehicles in the calibration image; The temporary ID and trajectory information of the motor vehicle are combined into motor vehicle data and cached in a corresponding matching pool.

4. The method according to claim 3, characterized in that The construction of the optimized YOLO algorithm model includes: Obtain a basic YOLO algorithm architecture, where the basic YOLO algorithm architecture includes a multi-scale feature fusion layer and a post-processing optimization module; A quarter-scale layer is added to the multi-scale feature fusion layer. The detailed features of the quarter-scale layer are fused through upsampling and the CSP double convolution bottleneck module. A space-to-depth conversion module is introduced to compress the spatial information into the depth channel, expand the receptive field, and retain the edge features of small objects. Embedding a selective kernel attention module in the multi-scale feature fusion layer and adding a coordinate attention module after the quarter-scale layer; In the post-processing optimization module, motion perception suppression is performed on motor vehicle identification through an adaptive non-maximum suppression mode, and noise filtering is performed through a multimodal trajectory filtering method.

5. The method according to claim 1, wherein The motor vehicle data in each matching pool is cyclically matched in a preset order, searching for motor vehicles appearing in multiple surveillance cameras, and saving the corresponding motor vehicle data to a result pool, including: Selecting a matching pool as a reference matching pool, cyclically matching the motor vehicle data in the reference matching pool with the motor vehicle data in other matching pools to determine whether the same motor vehicle is captured and identified by more than two surveillance cameras; If so, the motor vehicle data corresponding to the motor vehicle is placed in the result pool; if not, the motor vehicle data in the candidate pool is matched again. If the match is successful, the motor vehicle data corresponding to the motor vehicle is placed in the result pool; if the match is unsuccessful, the motor vehicle data is placed in the candidate pool; Each matching pool is configured with a corresponding candidate pool, and the candidate pool is used to store motor vehicle data that has not been successfully matched.

6. The method according to claim 5, characterized in that in, The determination of whether the same motor vehicle is photographed and identified by more than two surveillance cameras includes: During the matching process, a piece of motor vehicle data is extracted from each of the two matching pools, wherein the motor vehicle data includes a center coordinate value and a movement direction of the motor vehicle; Calculating the Euclidean distance according to the center coordinate values, and calculating the deflection angles of the two motor vehicles according to the movement directions; When the Euclidean distance and the deflection angle meet preset requirements, the two motor vehicles are determined to be the same physical entity, and a unique global ID is assigned to the physical entity.

7. The method according to claim 6, characterized in that The method further comprises: After the global ID is assigned, the motor vehicle data is saved to the result pool using a queue and hash table data structure; The motor vehicle data stored in the candidate pool is used to match the motor vehicle data in the next matching calculation cycle. If the match is successful, a global ID is assigned and placed in the result pool. If no match is successful within the preset time window, the invalid motor vehicle data is cleared.

8. The method according to claim 1, characterized in that The motor vehicle data includes the timestamp and speed change characteristics of the frame image, The determining whether the motor vehicle has a collision risk based on the time and speed change characteristics of the motor vehicle passing through the trajectory intersection includes: Obtain any trajectory intersection, filter all vehicles in the result pool that pass through the trajectory intersection, and record the global ID of the vehicle, the timestamp when it passes through the trajectory intersection, and the speed change characteristics; The time difference between multiple motor vehicles when passing through the trajectory intersection is calculated based on the timestamp. If the time difference is less than a preset time threshold, it is determined based on the speed change characteristics whether the acceleration of at least one motor vehicle is negative within a preset time range. If so, it is determined that there is a risk of collision between the motor vehicles at the trajectory intersection.

9. An electronic device, characterized in that: The invention comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for identifying conflicts between motor vehicles at an intersection as claimed in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for identifying conflicts between motor vehicles at an intersection according to any one of claims 1 to 8 is implemented.

Images