Overhead exit ramp connection area vehicle track restoration method, device and medium

Abstract

The invention discloses an elevated exit ramp connection area vehicle trajectory restoration method and device and a medium. The method comprises the following steps: determining video data cooperatively acquired by double unmanned aerial vehicles as double-view synchronous video data; performing video stitching and coordinate system unification based on the dual-view synchronous video data to generate a fused video; detecting the fused video through a YOLOv8 algorithm to obtain a vehicle position sequence, and performing state prediction on the vehicle position sequence through Kalman filtering to generate a motion state prediction result; performing trajectory reduction on the motion state prediction result and the position sequence to obtain a plurality of preliminary trajectory fragments of the vehicle; and combining the aligned preliminary trajectory fragments of the same vehicle through a Bayesian fusion algorithm, and restoring to generate a vehicle full trajectory chain. Through an improved SIFT algorithm, a YOLOv8 algorithm and the like, fusion and splicing of videos and accurate tracking of vehicle trajectories can be completed with high quality, and a more reliable data basis is provided for subsequent tasks such as vehicle trajectory extraction and traffic data analysis.

Description

Technical Field

[0001] This application relates to the field of traffic management, and in particular to a method, equipment and medium for reconstructing vehicle trajectories in the connection area of ​​elevated exit ramps. Background Technology

[0002] As a key node in the urban transportation network, the area connecting the exit ramps of urban elevated expressways and the ground-level main roads plays an important role in the conversion of elevated traffic flow and ground-level traffic flow. The traffic flow in this area is complex and intertwined, and vehicles frequently change lanes, which can easily lead to traffic conflicts, minor accidents, and regional congestion, seriously affecting traffic efficiency and safety.

[0003] Existing traffic monitoring methods in this area mainly rely on ground detection equipment such as inductive loops and fixed cameras. However, due to the obstruction caused by elevated bridge piers and bridge deck structures, these methods cannot cover the main road areas that are obstructed, resulting in incomplete trajectory data collection and inaccurate trajectory reconstruction.

[0004] Secondly, while existing single-drone monitoring methods have been applied in traffic monitoring due to their high-altitude perspective, they are limited by the field of view and cannot simultaneously cover the three core areas of elevated exit ramps, traffic weaving areas, and obscured main roads. Moreover, the traditional SIFT and YOLOv algorithms used in analyzing vehicle trajectories are easily affected by the environment in some complex road conditions and cannot stably capture vehicle image feature points, ultimately resulting in inaccurate vehicle trajectory reconstruction. Summary of the Invention

[0005] This application provides a method, device, and medium for vehicle trajectory restoration in the connection area of ​​elevated exit ramps, which solves the problems of incomplete monitoring view coverage and inaccurate vehicle trajectory restoration in existing vehicle trajectory restoration methods.

[0006] The embodiments of this application adopt the following technical solutions:

[0007] On one hand, this application provides a method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp. The method includes: determining video data synchronously collected by two UAVs as dual-view synchronous video data; performing feature matching on the dual-view synchronous video data using the SIFT algorithm and Euclidean distance to obtain effective matching point pairs; estimating the homography matrix of the effective matching point pairs using the adaptive RANSAC algorithm to spatially align the dual-view synchronous video data, and performing multi-band weighted fusion on the aligned dual-view synchronous video data to obtain a fused video; locating vehicles in the fused video using the YOLOv8 algorithm to obtain a vehicle position sequence, and predicting the vehicle's motion state between adjacent position points using extended Kalman filtering and the vehicle position sequence to obtain a vehicle motion state prediction result; performing bipartite graph matching between the vehicle motion state prediction result and the vehicle position sequence using the Hungarian algorithm to obtain multiple preliminary trajectory segments of the vehicle; aligning the multiple preliminary trajectory segments using timestamps, and merging the aligned multiple preliminary trajectory segments using a Bayesian fusion algorithm to generate the vehicle's full trajectory chain.

[0008] In one example, after merging multiple aligned preliminary trajectory segments using a Bayesian fusion algorithm to generate the vehicle's full trajectory chain, the method further includes: monitoring the vehicle's full trajectory chain using a dynamic window method to determine the vehicle's lateral displacement and motion state; determining whether the vehicle is performing a lane-changing event based on the lateral displacement and motion state; if so, marking the starting position of the lane-changing event when the vehicle completes the lane-changing event; counting the frequency of lane-changing events occurring within a circular area centered on the starting position and with a preset distance as the radius within a unit of time to determine the lane-changing frequency index; comparing a preset lane-changing frequency threshold with the lane-changing frequency index; if the lane-changing frequency index is greater than the lane-changing frequency threshold, marking the circular area as a frequent lane-changing area.

[0009] In one example, the determination of whether a vehicle is performing a lane-changing event is based on the lateral displacement and motion state. Specifically, the vehicle is determined to be performing a lane-changing event when the lateral displacement is greater than a preset displacement, the lateral velocity is greater than a preset lateral velocity, the longitudinal velocity is greater than a preset longitudinal velocity, the change in heading angle is greater than a preset angle change, and the duration of the heading angle change is greater than a preset duration.

[0010] In one example, feature matching is performed on the dual-view synchronized video data using the SIFT algorithm and Euclidean distance to obtain valid matching point pairs. Specifically, this includes: extracting feature points of video frames at the same time in the dual-view synchronized video data using the SIFT algorithm, and performing image gradient direction weighting processing on the feature points to obtain valid feature points; and performing bidirectional matching on the valid feature points using the Euclidean distance metric according to a preset distance threshold, and determining the feature point pairs that are successfully matched in both directions as valid feature point pairs.

[0011] In one example, the homography matrix of the effective matching point pairs is estimated using an adaptive RANSAC algorithm to spatially align the dual-view synchronized video data. Specifically, this includes: estimating the homography matrix of the effective matching point pairs using an adaptive RANSAC algorithm to determine the homography matrix of the dual-view synchronized video data; and performing a geometric transformation on the video frames acquired by the tilted UAV by adjusting the number of samplings based on the spatial transformation relationship in the homography matrix to spatially align the coordinate system of the tilted UAV video frames with that of the vertical UAV video frames.

[0012] In one example, the aligned dual-view synchronized video data is subjected to multi-band weighted fusion to obtain a fused video. Specifically, this includes: classifying the overlapping areas of the spatially aligned video frames to obtain low-frequency components and high-frequency components; performing weighted fusion on the low-frequency components to obtain a low-frequency fused video; performing gradient fusion on the high-frequency components to obtain a high-frequency fused video; and stitching the low-frequency fused video and the high-frequency fused video together to obtain a fused video.

[0013] In one example, vehicle location is performed on the fused video using the YOLOv8 algorithm to obtain a vehicle location sequence. Specifically, this includes: detecting vehicle elevated shadow features in the fused video using the YOLOv8 algorithm to obtain a first vehicle location result; enhancing the CBAM attention mechanism of the YOLOv8 algorithm using preset vehicle detection features, and detecting the first vehicle location result using the enhanced CBAM attention mechanism to obtain a vehicle location sequence.

[0014] In one example, the motion state of a vehicle between adjacent positions is predicted by using an extended Kalman filter and the vehicle position sequence to obtain the vehicle motion state prediction result. Specifically, this includes: performing linear state prediction processing on the vehicle position sequence using an extended Kalman filter based on a preset state vector to obtain the vehicle motion state prediction result; the linear state prediction processing includes predicting the center coordinates of the vehicle detection box and predicting the longitudinal and lateral velocities of the vehicle.

[0015] On the other hand, embodiments of this application provide a vehicle trajectory restoration device for elevated exit ramp connection areas, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any of the above-mentioned elevated exit ramp connection area vehicle trajectory restoration methods.

[0016] On the other hand, embodiments of this application provide a non-volatile computer storage medium for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp, storing computer-executable instructions that can execute any of the above-mentioned methods for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp.

[0017] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects:

[0018] This application utilizes dual drones for collaborative filming, enabling comprehensive acquisition of traffic data from the interchange areas of urban elevated expressway exit ramps, thus improving the accuracy and completeness of traffic monitoring. Through improved SIFT and other algorithms, high-quality fusion and stitching of the captured videos can be achieved, providing a more reliable data foundation for subsequent vehicle trajectory extraction and traffic data analysis. Improved YOLOv8 and other algorithms enable more accurate vehicle trajectory tracking in more complex traffic scenarios, providing a more effective means to analyze traffic flow characteristics and assess traffic operation status. The identified areas with frequent lane changes can provide a scientific basis for traffic management departments to formulate reasonable traffic management strategies, contributing to improved traffic safety and operational efficiency in these areas. Attached Figure Description

[0019] To more clearly illustrate the technical solution of this application, some embodiments of this application will be described in detail below with reference to the accompanying drawings, in which:

[0020] Figure 1 A flowchart illustrating a method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp, provided in an embodiment of this application;

[0021] Figure 2 A system architecture diagram of a method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp provided in this application embodiment;

[0022] Figure 3 A schematic diagram of a vertical UAV image used to illustrate a method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp, as provided in an embodiment of this application.

[0023] Figure 4 A tilting UAV photograph is used to illustrate a method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp, as provided in an embodiment of this application.

[0024] Figure 5 This is a schematic diagram of a vehicle trajectory restoration device for the connection area of ​​an elevated exit ramp, provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Some embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0027] Figure 1 This is a flowchart illustrating a method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp, as provided in this embodiment of the application. This method can be applied to different business domains. Certain input parameters or intermediate results in this process can be manually adjusted to help improve accuracy.

[0028] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a controller as an example.

[0029] Based on this Figure 1 The process may include the following steps:

[0030] S101: The video data collected collaboratively by two drones is identified as dual-view synchronous video data.

[0031] In some embodiments of this application, before conducting collaborative data collection using two drones, it is necessary to configure the two drones. Specifically, the first high-altitude vertical-view drone, equipped with a high-definition camera, flies to a preset altitude and captures the overall scene of the connecting area from a 90° vertical perspective, covering the elevated exit ramp, the unobstructed area of ​​the main road, and the weaving area, to obtain video data containing the macroscopic movement trajectory of vehicles. The second low-altitude tilt-view drone, equipped with the same high-definition camera, flies to a preset low altitude and captures the main road area obstructed by the elevated road from a specific tilt angle, focusing on the microscopic movement state of vehicles in the obstructed road section to ensure the capture of trajectory segments of vehicles entering / exiting the obstructed area.

[0032] Both drones started filming at the same time, with the video frame rate uniformly set to 30fps to ensure consistent timestamps.

[0033] By using two drones for collaborative shooting, comprehensive traffic data can be obtained from the connection areas of the exit ramps of urban elevated expressways, improving the accuracy and completeness of traffic monitoring.

[0034] S102: Using the SIFT algorithm and Euclidean distance, feature matching is performed on the dual-view synchronized video data to obtain effective matching point pairs.

[0035] In some embodiments of this application, after acquiring dual-view synchronized video data, feature points are extracted from the dual-view synchronized video data. For video frames captured by two drones at the same moment, feature points are extracted using the Scale Invariant Feature Transform (SIFT) algorithm. The SIFT algorithm can stably extract feature points from images under different scales, rotations, and lighting conditions. Its calculation formula is as follows:

[0036] L(x p y p ,σ)=G(x p y p ,σ)*I(x p y p )

[0037] Where, x p It is the horizontal pixel coordinate of the feature point in the image, y p L(x) represents the vertical pixel coordinates of the feature point in the image, σ is the scale parameter, and L(x) represents the vertical pixel coordinates of the feature point in the image. p y p G(x) is the difference-of-Gaussian pyramid (DOG) image. p y p σ) is the Gaussian kernel function, I(x) p y p The image shown is the original image. A DOG image is obtained by subtracting Gaussian blurred images at different scales, and extreme points are detected on the DOG image as feature points.

[0038] When using the SIFT algorithm to extract feature points, the number of scale-space layers is set to 8, with 3 scales per group, and the number of feature points is controlled at 1000-1500 per frame. Furthermore, to further improve the robustness of feature point extraction, an improved SIFT algorithm is introduced. In the feature point descriptor generation stage, a weighted processing of the image gradient direction is added, with the specific formula as follows:

[0039]

[0040] Where m(x, y) is the gradient magnitude and θ(x, y) is the gradient direction. By weighting the gradient direction, the adaptability of feature points to changes in illumination is enhanced.

[0041] Furthermore, bidirectional matching of effective feature points is performed using the Euclidean distance metric. For each feature point p1(x1, y1) in the vertical UAV video frame, the nearest feature point p2(x2, y2) in the tilted UAV video frame is found. The distance calculation formula is as follows:

[0042]

[0043] To improve the accuracy of matching, a distance threshold T is set. Only when d(p1, p2) < T, these two feature points are considered to be successfully matched. At the same time, a two-way matching verification mechanism is introduced, that is, not only looking for matching points from the feature points of the vertical drone to the tilted drone, but also looking for matching points from the feature points of the tilted drone to the vertical drone. Only when both directions are successfully matched, it is confirmed as a valid matching point pair, further improving the accuracy of matching.

[0044] S103: Estimate the homography matrix for the valid matching point pairs through the adaptive RANSAC algorithm to spatially align the dual-view synchronized video data, and perform multi-band weighted fusion on the aligned dual-view synchronized video data to obtain a fused video.

[0045] In some embodiments of the present application, after obtaining the effective feature matching point pairs, according to the successfully matched feature point pairs, use the RANSAC algorithm to estimate the homography matrix B from the matching point pairs. By randomly sampling 1000 times and an inlier threshold of 3 pixels, transform the low-altitude tilted view video frame into the high-altitude vertical view coordinate system to achieve the spatial alignment of the two videos. To improve the efficiency of the RANSAC algorithm, an adaptive sampling number calculation method is introduced. According to the number of initial matching point pairs and the expected inlier rate, dynamically adjust the sampling number. The formula is as follows:

[0046]

[0047] where N is the sampling number, ε is the confidence level, γ is the expected inlier rate, and m is the number of points sampled each time. By adaptively adjusting the sampling number, while ensuring the accuracy, improve the algorithm efficiency.

[0048] Furthermore, for the overlapping area, use fade-in and fade-out weighted fusion. The pixel weight ω is calculated according to its distance d from the splicing boundary: [[ID=]]

[0049]

[0050] where λ is the weight decay coefficient to ensure that the spliced video is smooth and has no cracks. The width of the overlapping area is set to 100 pixels, and the weight decay coefficient λ = 0.15. The weighted fusion is implemented through the Python OpenCV library. At the same time, adopt a multi-band fusion strategy, decompose the image into low-frequency and high-frequency components, use the weighted fusion method for the low-frequency components, and use the gradient-based fusion method for the high-frequency components to improve the quality of the spliced video.

[0051] S104: The vehicle is located in the fused video using the YOLOv8 algorithm to obtain a vehicle position sequence. The vehicle's motion state between adjacent positions is predicted by using an extended Kalman filter and the vehicle position sequence to obtain the vehicle motion state prediction result.

[0052] In some embodiments of this application, after the videos of two drones are stitched and fused, an improved YOLOv8 algorithm is used to detect vehicles in the fused video. An elevated road shadow feature extraction branch is added to the YOLOv8 model, and an attention mechanism is embedded in the backbone network to improve vehicle detection accuracy. Specifically, the elevated road shadow feature extraction branch adopts a convolutional neural network structure, specifically designed to extract features of elevated road shadows. It is trained jointly with the vehicle detection task through multi-task learning, enabling the model to better distinguish between vehicles and elevated road shadows, reducing false detections. The attention mechanism uses CBAM (Convolutional Block Attention Module), including channel attention and spatial attention, which can automatically learn features more important for vehicle detection, suppress irrelevant features, and improve detection accuracy. Simultaneously, COCO pre-trained weights are used, the detection confidence threshold is set to 0.6, and vehicle category filtering ensures that only motor vehicle trajectories are retained.

[0053] Furthermore, by combining Kalman filtering, state prediction is performed on the detected vehicle positioning sequence, using the following formula:

[0054] S k|k-1 =AS k-1|k-1 +ω k-1

[0055] Z k =HS k|k-1 +V k

[0056] Wherein, the state vector S contains vehicle coordinates and velocity, k is the time step index, representing the k-th time in the discrete time series, A is the state transition matrix, H is the observation matrix, and ω k-1 and V k This refers to process noise and observation noise.

[0057] To improve the adaptability of Kalman filtering to nonlinear motion, an extended Kalman filter (EKF) is introduced. This linearizes the state transition equations and observation equations. The extended Kalman filter state vector s = [x...] c y c v x v y ] T x c Let y be the horizontal pixel coordinate of the center of the vehicle detection box. cv is the vertical pixel coordinate of the center of the vehicle detection box. x v is the longitudinal speed of the vehicle along the road. y Let $\frac{ ... The observation matrix H is taken as the coordinates of the center of the detection box. Extended Kalman filtering can better handle nonlinear motions such as vehicle acceleration, deceleration, and turning.

[0058] S105: The vehicle motion state prediction result and the vehicle position sequence are matched by bipartite graph using the Hungarian algorithm to obtain multiple preliminary trajectory segments of the vehicle; the bipartite graph matching uses Mahalanobis distance and appearance feature distance as matching costs.

[0059] In some embodiments of this application, the Hungarian algorithm is used to solve the bipartite graph matching problem between the predicted trajectory and the detection box, with Mahalanobis distance and appearance feature distance as the matching cost:

[0060] C(i,j)=αd m (i, j)+(1-α)d f (i, j)

[0061] Where, d m Let d be the Mahalanobis distance. f α is the feature distance, and α is the weight coefficient.

[0062] The Hungarian algorithm uses a matching period of 10 frames. Appearance features are extracted as 128-dimensional vectors using ResNet-18, with Mahalanobis distance and feature distance weights α = 0.7. Spatiotemporal constraints are introduced, requiring the matched trajectories to satisfy both temporal and spatial continuity to reduce false matches.

[0063] S106: Align the multiple preliminary trajectory segments using timestamps, and merge the aligned preliminary trajectory segments using a Bayesian fusion algorithm to generate the vehicle's full trajectory chain.

[0064] In some embodiments of this application, after reconstructing the trajectory of the motion state prediction results using the Hungarian algorithm to obtain multiple preliminary trajectory segments of the vehicle, it is necessary to align the trajectory segments of two UAVs using timestamps, and then use a Bayesian fusion algorithm to merge the trajectories of the same vehicle from different perspectives to generate a complete trajectory chain, thereby analyzing areas with frequent lane changes. The Bayesian fusion algorithm, based on Bayesian estimation theory, uses trajectories from different perspectives as observations, updates the prior probability distribution to obtain the posterior probability distribution, and thus obtains a more accurate trajectory estimate. Specifically, for the trajectories u1 and u2 of the same vehicle from different perspectives, the fused trajectory... The estimated value is:

[0065]

[0066] in, and For trajectory segments u1 and u2 in the fused trajectory Likelihood probability under given conditions For fusion trajectory The prior probabilities are given by p(u1, u2), which is the joint probability of trajectory segments u1 and u2. By employing the Bayesian fusion algorithm, trajectory information from different perspectives can be fully utilized, improving the accuracy and completeness of the trajectory.

[0067] Furthermore, after obtaining the vehicle's full trajectory chain, the dynamic window method is used to monitor the vehicle's full trajectory chain and determine the vehicle's lateral displacement and motion state. The motion state includes lateral velocity, longitudinal velocity, yaw angle change, and yaw angle change duration. Then, based on the lateral displacement and motion state, it is determined whether the vehicle is performing a lane change event. That is, when the lateral displacement is greater than a preset displacement, the lateral velocity is greater than a preset lateral velocity, the longitudinal velocity is greater than a preset longitudinal velocity, the yaw angle change is greater than a preset angle change, and the yaw angle change duration is greater than a preset duration, it is determined that the vehicle is performing a lane change event.

[0068] Furthermore, if a vehicle completes a lane-change event, the location of the determined lane-change event is marked. On the interface map, the starting position of each lane-change event is presented as a marker. The number of lane changes N within a certain range around each marker position per unit time is counted, and the lane-change frequency index F is defined.

[0069]

[0070] Where M represents the area of ​​the grid within the statistical range surrounding the punctuation mark, and t represents the statistical time.

[0071] Set the preset lane change frequency threshold (e.g., 0.02 times / (m)). 2 The lane change frequency index is compared with the lane change frequency index. If the lane change frequency index is greater than the lane change frequency threshold, the range corresponding to the lane change frequency index is marked as a frequent lane change area and a regional heat map is generated.

[0072] This invention, through the analysis of vehicle trajectory data, can accurately identify areas with frequent lane changes at the junction of urban elevated expressway exit ramps and urban main roads, providing a scientific basis for traffic management departments to formulate reasonable traffic management strategies and helping to improve traffic safety and operational efficiency in these areas.

[0073] It should be noted that, although the embodiments in this application are based on... Figure 1Steps S101 to S106 will be described sequentially, but this does not mean that steps S101 and S106 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S101 to S106 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S101 to S106 can be appropriately adjusted according to actual needs.

[0074] pass Figure 1 This application employs a method involving collaborative filming with two drones to comprehensively acquire traffic data from the interchange areas of urban elevated expressway exit ramps, improving the accuracy and completeness of traffic monitoring. Through improved SIFT and other algorithms, the captured videos can be fused and stitched together with high quality, providing a more reliable data foundation for subsequent vehicle trajectory extraction and traffic data analysis. Improved YOLOv8 and other algorithms enable more accurate vehicle trajectory tracking in more complex traffic scenarios, providing a more effective means to analyze traffic flow characteristics and assess traffic operation status. The finally identified areas with frequent lane changes can provide a scientific basis for traffic management departments to formulate reasonable traffic management strategies, helping to improve traffic safety and operational efficiency in these areas.

[0075] Figure 2 This is a system architecture diagram of a method for restoring vehicle trajectory in the connection area of ​​an elevated exit ramp, provided in an embodiment of this application.

[0076] exist Figure 2 The paper presents the overall framework of this application, from dual-drone shooting to video timestamp synchronization, to improved SIFT feature matching, to adaptive RANSAC stitching, to improved YOLOv8 detection, to extended Kalman filtering, to Bayesian trajectory fusion, to generating the full trajectory chain, and finally to generating heatmaps of areas with frequent lane changes.

[0077] Figure 3 This is a schematic diagram taken by a vertical UAV for a method of restoring vehicle trajectory in the connection area of ​​an elevated exit ramp provided in an embodiment of this application.

[0078] exist Figure 3 The image shows the specific shooting position and shooting angle of the vertical drone in this application.

[0079] Figure 4 This is a schematic diagram taken by a tilting UAV, illustrating a method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp provided in this application embodiment.

[0080] exist Figure 4The image shows the specific shooting position and shooting angle of the tilting drone used in this application.

[0081] Figure 5 A schematic diagram of a vehicle trajectory reconstruction device for an elevated exit ramp connection area provided in this application embodiment includes:

[0082] At least one processor; and,

[0083] A memory that is communicatively connected to at least one processor; wherein,

[0084] A method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp, wherein the memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable at least one processor to perform any of the above-mentioned tasks.

[0085] Some embodiments of this application provide a non-volatile computer storage medium for vehicle trajectory reconstruction in the connection area of ​​an elevated exit ramp, which stores computer-executable instructions capable of executing any of the above-mentioned methods for vehicle trajectory reconstruction in the connection area of ​​an elevated exit ramp.

[0086] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0087] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0088] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0089] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0090] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0091] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0092] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0093] Memory may include non-persistent storage in computer-readable media, random access memory (RAM), and non-volatile memory such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0094] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0095] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0096] The above are merely embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the technical principles of this application should fall within the protection scope of this application.

Claims

1. A method for reconstructing vehicle trajectories in the connection area of ​​an elevated exit ramp, characterized in that, The method includes: The video data collected synchronously by the two drones is defined as dual-view synchronous video data; the vertical drone in the two drones shoots the overall scene of the connecting area from a vertical top view at a preset height, and the tilting drone shoots the main road area that is blocked by the overpass at a preset tilt angle. By using the SIFT algorithm and Euclidean distance, feature matching is performed on the dual-view synchronized video data to obtain effective matching point pairs; The homography matrix of the effective matching point pairs is estimated by the adaptive RANSAC algorithm to spatially align the dual-view synchronous video data, and the aligned dual-view synchronous video data is then fused by multi-band weighted fusion to obtain the fused video. The YOLOv8 algorithm is used to locate vehicles in the fused video to obtain a vehicle position sequence. Then, the motion state of the vehicle between adjacent position points is predicted by using an extended Kalman filter and the vehicle position sequence to obtain the vehicle motion state prediction result. The vehicle motion state prediction results and vehicle position sequence are matched using the Hungarian algorithm to obtain multiple preliminary trajectory segments of the vehicle; the bipartite graph matching uses Mahalanobis distance and appearance feature distance as matching costs. The multiple preliminary trajectory segments are aligned using timestamps, and the aligned preliminary trajectory segments are merged using a Bayesian fusion algorithm to generate the vehicle's full trajectory chain.

2. The method according to claim 1, characterized in that, After merging the aligned preliminary trajectory segments using a Bayesian fusion algorithm to generate the vehicle's full trajectory chain, the method further includes: The vehicle's entire trajectory chain is monitored using a dynamic window method to determine the vehicle's lateral displacement and motion state; the motion state includes lateral velocity, longitudinal velocity, change in heading angle, and duration of the heading angle change. Based on the lateral displacement and motion state, determine whether the vehicle has performed a lane-changing event; If so, mark the starting point of the lane change when the vehicle completes the lane change event; The frequency of lane changes by vehicles within a circular area centered at the starting point and with a preset distance as the radius is counted per unit time to determine the lane change frequency index. The preset lane-changing frequency threshold is compared with the lane-changing frequency index. If the lane-changing frequency index is greater than the lane-changing frequency threshold, the circular area is marked as a frequent lane-changing area.

3. The method according to claim 2, characterized in that, The method of determining whether a vehicle has initiated a lane-changing event based on lateral displacement and motion state specifically includes: When the lateral displacement is greater than the preset displacement, the lateral velocity is greater than the preset lateral velocity, the longitudinal velocity is greater than the preset longitudinal velocity, the change in heading angle is greater than the preset angle change, and the duration of the heading angle change is greater than the preset duration, it is determined that the vehicle is in the process of changing lanes.

4. The method according to claim 1, characterized in that, The step of performing feature matching on the dual-view synchronized video data using the SIFT algorithm and Euclidean distance to obtain effective matching point pairs specifically includes: Feature points of video frames at the same time in dual-view synchronized video data are extracted using the SIFT algorithm, and the feature points are weighted by the image gradient direction to obtain effective feature points. Based on a preset distance threshold, the effective feature points are bidirectionally matched using the Euclidean distance metric, and feature point pairs that are successfully matched in both directions are determined as effective feature point pairs.

5. The method according to claim 1, characterized in that, The step of estimating the homography matrix of the effective matching point pairs using the adaptive RANSAC algorithm to spatially align the dual-view synchronized video data specifically includes: The homography matrix of the dual-view synchronized video data is determined by estimating the homography matrix of the effective matching point pairs using the adaptive RANSAC algorithm. Based on the spatial transformation relationship in the homography matrix, the video frames acquired by the tilted UAV are geometrically transformed by adjusting the number of samplings in order to spatially align the coordinate system of the tilted UAV video frames with that of the vertical UAV video frames.

6. The method according to claim 1, characterized in that, The aligned dual-view synchronized video data is then subjected to multi-band weighted fusion to obtain the fused video, which specifically includes: The overlapping regions of spatially aligned video frames are classified to obtain low-frequency and high-frequency components. The low-frequency components are weighted and fused to obtain a low-frequency fused video, and the high-frequency components are gradient fused to obtain a high-frequency fused video. The low-frequency fused video and the high-frequency fused video are then stitched together to obtain a fused video.

7. The method according to claim 1, characterized in that, The step of performing vehicle localization on the fused video using the YOLOv8 algorithm to obtain a vehicle location sequence specifically includes: The YOLOv8 algorithm is used to detect the elevated road shadow features of the fused video to obtain the first vehicle localization result; the purpose of the first vehicle localization result is to distinguish between vehicles and elevated road shadows in the fused video. The CBAM attention mechanism of the YOLOv8 algorithm is enhanced by pre-defined vehicle detection features, and the enhanced CBAM attention mechanism is used to detect the first vehicle localization result to obtain the vehicle position sequence.

8. The method according to claim 1, characterized in that, By using an extended Kalman filter and the vehicle position sequence, the motion state of the vehicle between adjacent position points is predicted, resulting in a vehicle motion state prediction. Specifically, this includes: Based on a preset state vector, the vehicle position sequence is subjected to linear state prediction processing through extended Kalman filtering to obtain the vehicle motion state prediction result; the linear state prediction processing includes prediction of the center coordinates of the vehicle detection box and prediction of the vehicle's longitudinal and lateral velocities.

9. A vehicle trajectory reconstruction device for the connection area of ​​an elevated exit ramp, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the vehicle trajectory restoration method for the elevated exit ramp connection area as described in any one of claims 1-8.

10. A vehicle trajectory reconstruction storage medium for elevated exit ramp connection areas, storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of executing the vehicle trajectory restoration method for the elevated exit ramp connection area as described in any one of claims 1-8.

Images