Vehicle driving area generation method and system

Abstract

The invention relates to the technical field of intelligent traffic, in particular to a vehicle driving area generation method and system, and the method comprises the steps: obtaining a vehicle position coordinate in a traffic image, and constructing a time-space point cloud; carrying out outlier filtering on the space-time point clouds, carrying out lane clustering based on the filtered space-time point clouds, and separating out point cloud clusters representing different lanes; calculating an alpha value based on the space density of the space-time point cloud; inputting the alpha value into an Alpha Shape algorithm to generate a road network level driving area, and independently applying the alpha value to each point cloud cluster to generate a lane level driving area; and generating a dynamic obstacle area based on the filtered static vehicle position points, integrating the generated road network level area and the lane level driving area, and outputting a vehicle multi-level driving area. According to the invention, the problems that a real dynamic driving area is difficult to accurately extract and abnormal tracks and outliers are insufficient to process in the prior art are effectively solved, and the accuracy of driving area generation and the adaptability to a complex traffic environment are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method and system for generating vehicle driving areas. Background Technology

[0002] Currently, vehicle driving area extraction mainly relies on high-precision maps, road marking detection, or rule-based image segmentation methods. These methods are mostly based on static geometric information, usually requiring clear and complete road markings, relying on stable prior map data, and mainly using two-dimensional image information for region inference. They lack in-depth utilization of the actual movement behavior of vehicles. With the increasing complexity of traffic environments, there are often situations such as construction, worn road markings, temporary detours, or traffic congestion on actual roads, making it difficult for traditional methods to adapt to dynamic, complex, and unstructured traffic scenarios.

[0003] Against this backdrop, existing methods struggle to accurately extract the vehicle's actual dynamic driving area, fail to fully reflect the vehicle's trajectory and real-time behavioral characteristics, and lack the ability to handle abnormal trajectories and outliers. This results in significant discrepancies between the extracted results and actual driving conditions, limiting their reliability and practicality in applications such as autonomous driving path planning, traffic management, and high-precision map updates. Summary of the Invention

[0004] This invention provides a method and system for generating vehicle driving areas, thereby effectively solving the problems pointed out in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for generating a vehicle driving area, comprising:

[0007] Obtain vehicle location coordinates from traffic images and construct a spatiotemporal point cloud;

[0008] Outlier filtering is performed on the spatiotemporal point cloud, and lane clustering is performed based on the filtered spatiotemporal point cloud to separate point cloud clusters representing different lanes.

[0009] The value of α is calculated based on the spatial density of the spatiotemporal point cloud. The value of α is the maximum radius of the circumcircle of the generated polygon as defined in the Alpha Shape algorithm.

[0010] The α value is input into the Alpha Shape algorithm to generate a road network-level driving area, and the α value is applied independently to each point cloud cluster to generate a lane-level driving area;

[0011] Dynamic obstacle zones are generated based on the filtered stationary vehicle location points. The generated road network-level areas and lane-level driving areas are integrated to output multi-level vehicle driving areas.

[0012] Further, the calculation of the α value based on the spatial density of the spatiotemporal point cloud includes:

[0013] The filtered spatiotemporal point cloud is divided into equal-sized grid units;

[0014] Calculate the number of points in each of the grid cells, and determine the local α value based on the inverse relationship of the number of points.

[0015] The local α values ​​are sorted, and the values ​​at specific quantile positions in the sort are selected as the global α values.

[0016] Furthermore, the value at the specific quantile position is not less than half the standard lane width.

[0017] Further, outlier filtering is performed on the spatiotemporal point cloud, including:

[0018] Detect stationary vehicle point clouds in the spatiotemporal point cloud whose velocity is continuously below a set threshold, and identify associated point clouds that have spatial continuity with the stationary vehicle point clouds;

[0019] Isolated points that are discontinuously distributed and have a velocity exceeding a threshold are marked as outliers;

[0020] The filtering radius is adjusted based on the spatial density distribution of the outliers, and all outliers within the filtering radius are removed.

[0021] Further, adjusting the filtering radius based on the spatial density distribution of the outliers includes:

[0022] A radial distance function is established with the outlier point as the center, and the variance of the point cloud density within a preset range is calculated;

[0023] When the variance value exceeds the standard deviation threshold, the filtering radius is increased proportionally.

[0024] When the variance value does not exceed the standard deviation threshold, an initial fixed radius value is used for filtering.

[0025] Furthermore, vehicle location coordinates are collected from traffic images and a spatiotemporal point cloud is constructed, including:

[0026] Vehicle position coordinates are extracted from the traffic image using an object detection algorithm;

[0027] The extracted vehicle location coordinates are converted into three-dimensional spatiotemporal coordinates, where two dimensions represent the geographic spatial location and the third dimension represents the timestamp;

[0028] The spatiotemporal coordinates of the same vehicle within a continuous time period are associated to form a motion trajectory, and the motion trajectory data of all vehicles are aggregated to generate the spatiotemporal point cloud.

[0029] Further, the α value is input into the Alpha Shape algorithm to generate a road network-level driving region, and the α value is independently applied to each point cloud cluster to generate a lane-level driving region, including:

[0030] The global α value is input into the Alpha Shape algorithm to reconstruct the surface of the filtered complete spatiotemporal point cloud, generating a continuous driving area covering the entire road network;

[0031] Based on the parallelism metric of vehicle motion vectors, spatiotemporal point clouds are segmented into independent point cloud clusters;

[0032] Input the same α value as the road network level to each point cloud cluster, and generate lane-level driving areas using the Alpha Shape algorithm.

[0033] Furthermore, a dynamic obstacle zone is generated based on the filtered stationary vehicle location points. The generated road network-level region and lane-level driving region are integrated to output a multi-level vehicle driving region, including:

[0034] Dynamic obstacle avoidance zones are generated based on static obstacle point clouds. When these zones intersect with the original lane-level area space, lane-level continuous navigable channels are reconstructed in real time.

[0035] The road network-level area, lane-level continuous navigable channel, and dynamic obstacle avoidance zone are integrated into a three-layer structure to output the vehicle's multi-level driving area.

[0036] A vehicle driving area generation system, the system comprising:

[0037] The point cloud construction module acquires the vehicle location coordinates in traffic images and constructs a spatiotemporal point cloud.

[0038] The lane clustering module filters outliers from the spatiotemporal point cloud and performs lane clustering based on the filtered spatiotemporal point cloud to separate point cloud clusters that represent different lanes.

[0039] The α value calculation module calculates the α value based on the spatial density of the spatiotemporal point cloud. The α value is the maximum radius of the circumcircle of the generated polygon as defined in the Alpha Shape algorithm.

[0040] The α value application module inputs the α value into the Alpha Shape algorithm to generate a road network-level driving area, and independently applies the α value to each point cloud cluster to generate a lane-level driving area;

[0041] The region output module generates a dynamic obstacle zone based on the filtered stationary vehicle location points, integrates the generated road network-level region and lane-level region, and outputs a multi-level driving region for vehicles.

[0042] Furthermore, the α value application module includes:

[0043] The continuous region generation unit inputs the global α value into the Alpha Shape algorithm to perform surface reconstruction on the filtered complete spatiotemporal point cloud, generating a continuous driving region covering the entire road network;

[0044] The spatiotemporal point cloud segmentation unit, based on the parallelism metric of vehicle motion vectors, segments the spatiotemporal point cloud into independent point cloud clusters;

[0045] The lane-level region generation unit inputs the same α value as the road network-level value to each of the point cloud clusters and generates lane-level driving regions using the AlphaShape algorithm.

[0046] The technical solution of this invention can achieve the following technical effects:

[0047] It effectively solves the problems of existing methods that are difficult to accurately extract real dynamic driving areas and are insufficient in handling abnormal trajectories and outliers, thus improving the accuracy of driving area generation and adaptability to complex traffic environments. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating a method for generating a vehicle driving area.

[0050] Figure 2 This is a flowchart illustrating the process of calculating the α value based on the spatial density of spatiotemporal point clouds.

[0051] Figure 3 This is a flowchart illustrating the process of filtering outliers in a spatiotemporal point cloud.

[0052] Figure 4 A schematic diagram of the process for adjusting the filter radius based on the spatial density distribution of outliers;

[0053] Figure 5 A flowchart illustrating the process of obtaining vehicle location coordinates from traffic images and constructing a spatiotemporal point cloud;

[0054] Figure 6 A flowchart illustrating the process of generating road network-level and lane-level driving areas;

[0055] Figure 7A flowchart illustrating the process of outputting multi-level driving areas for vehicles. Detailed Implementation

[0056] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0058] Example 1

[0059] like Figure 1 As shown, the present invention provides a method for generating a vehicle driving area, the method comprising:

[0060] S1: Obtain the vehicle location coordinates in traffic images and construct a spatiotemporal point cloud;

[0061] Specifically, it is preferable to use continuous image sequences collected by fixed traffic monitoring cameras deployed in areas such as road intersections, main roads, and entrances and exits as input data. Mobile monitoring or drone aerial images can also be used as supplementary data sources to obtain more comprehensive vehicle location information. During image processing, it is preferable to use a target detection algorithm based on convolutional neural networks (such as the YOLO series algorithms) to automatically detect the vehicle position in each frame of the image and output the two-dimensional position coordinates of the vehicle in the image coordinate system. Time alignment is also performed, and each point contains the actual two-dimensional position of the vehicle at a certain time and the corresponding time information. By aggregating the three-dimensional spatiotemporal coordinates of all vehicles at different time points, a complete spatiotemporal point cloud can be constructed.

[0062] S2: Filter outliers from the spatiotemporal point cloud and perform lane clustering based on the filtered spatiotemporal point cloud to separate point cloud clusters that represent different lanes.

[0063] Specifically, firstly, for the constructed spatiotemporal point cloud, outliers are identified by analyzing the velocity information of each point and the trajectory continuity within adjacent time periods. Preferably, a velocity threshold and a direction change threshold can be set based on the vehicle's velocity and direction change rate in continuous time frames. When the velocity of some points is much higher than normal traffic flow or the trajectory changes abruptly, these points can be marked as outliers. In addition, spatial neighborhood density analysis can be used to identify points that significantly deviate from the mainstream trajectory and are spatially isolated, and these can be regarded as outliers. To further improve the filtering accuracy, it is preferable to combine statistical methods to judge the distribution characteristics of outliers and remove them according to preset spatial distance thresholds or neighborhood point number thresholds, ultimately obtaining cleaner point cloud data.

[0064] After outlier filtering, lane clustering is performed based on the filtered high-quality spatiotemporal point cloud. By analyzing the spatial distribution density of the point cloud and the consistency of vehicle movement direction, the vehicle trajectories are automatically grouped. In this process, the vehicle's motion vector is used as a key feature to assist clustering, so as to ensure that the trajectory direction of vehicles in the same lane is highly consistent and to avoid vehicles from different lanes from being mixed into the same cluster. The clustering result divides the spatiotemporal point cloud into multiple point cloud clusters, each cluster corresponding to an actual existing lane, thereby achieving lane-level spatial separation.

[0065] S3: Calculate the α value based on the spatial density of spatiotemporal point clouds. The α value is the maximum radius of the circumcircle of the generated polygon as defined in the Alpha Shape algorithm.

[0066] Specifically, the α value, as a control parameter defined in the Alpha Shape algorithm, is used to determine the maximum radius of the circumcircle of the generated polygon. Its key role is to adjust the fitting degree of the region boundary. The driving trajectory of vehicles on actual roads exhibits significant dynamism and diversity. Especially in complex scenarios such as missing lane markings, lane adjustments, road construction, or traffic congestion, the spatial distribution of vehicles is often irregular and uneven in density. Traditional methods are difficult to reliably extract accurate boundaries. In this context, the Alpha Shape algorithm introduces the α value, which can flexibly control the tightness of the boundary curve according to the distribution characteristics of the point cloud in different regions: when the α value is large, the generated contour is smoother, which is suitable for the overall modeling of sparse regions; when the α value is small, it can fit the edge of the point cloud more meticulously, which is suitable for the boundary restoration of dense or complex-shaped regions.

[0067] S4: Input the α value into the Alpha Shape algorithm to generate the road network-level driving area, and apply the α value independently to each point cloud cluster to generate the lane-level driving area;

[0068] Specifically, by introducing the Alpha Shape algorithm, the calculated α value is applied to both the overall point cloud and the point cloud clusters of each lane, achieving the reconstruction of driving area boundaries from macro to micro levels. This process aims to generate a continuous road network-level driving area that covers the entire road network and accurately represents the actual travel range of each lane.

[0069] S5: Generates dynamic obstacle zones based on the filtered stationary vehicle location points, integrates the generated road network-level areas and lane-level driving areas, and outputs multi-level driving areas for vehicles.

[0070] Specifically, by identifying the location points of filtered stationary vehicles, a dynamic obstacle avoidance zone is generated to reflect the actual traffic obstruction caused by temporary parking, congestion, or sudden events on the road. Based on this, the original lane-level area is corrected in real time to construct a continuous navigable channel. Finally, the road network-level area, lane-level channel, and dynamic obstacle zone are integrated into a unified multi-level structure, thereby improving the adaptability of the driving area model to complex traffic environments and its responsiveness to dynamic changes, ensuring that the output results are closer to the real traffic conditions.

[0071] As a preferred embodiment of the above, such as Figure 2 As shown, the calculation of the α value based on the spatial density of spatiotemporal point clouds includes:

[0072] A10: Divide the filtered spatiotemporal point cloud into equal-sized grid units;

[0073] A20: Calculate the number of points in each grid cell and determine the local α value based on the inverse relationship of the number of points.

[0074] A30: Sort the local α values ​​and select the value at a specific quantile position in the sort as the global α value.

[0075] Specifically, the high-quality spatiotemporal point cloud, filtered for outliers, is first projected onto the geographic plane, and the entire region is divided into uniformly sized grid units. The grid size can be set according to the actual road width, average vehicle distribution density, and accuracy requirements, such as 0.5m × 0.5m or 1m × 1m. In this preferred approach, by reasonably setting the grid size, both computational efficiency and local detail representation can be balanced, avoiding excessively large grids that lead to feature blurring or excessively small grids that increase the processing burden. The number of point clouds in each grid unit is counted to obtain the spatial point density of each grid. Then, based on the characteristics of vehicle trajectory distribution, the reciprocal relationship of the number of point clouds is preferred to calculate the local α value. That is, the higher the point cloud density, the smaller the corresponding local α value, making the boundary fitting of the region more compact and accurate. Conversely, in areas with sparse point clouds, the local α value is relatively large to ensure that the generated region is not excessively fragmented or broken, thus maintaining overall coherence. During this process, local α values ​​can be smoothed, for example, by considering the average or weighted average of adjacent grid cells, to further improve the stability of local α values. All calculated local α values ​​are then sorted as a whole, and a representative global α value is selected based on the set quantile strategy. For example, the median (50th percentile) or a higher quantile (such as the 75th percentile) can be selected as the global α value to ensure the continuity of the overall area while also taking into account the depiction of local details. When there are obvious main roads and auxiliary roads, merging areas, or congested areas in the road, selecting a mid-to-high quantile α value can more effectively balance the dense trajectory of the main road and the sparse trajectory of the auxiliary road, ensuring that the final generated overall driving area is neither coarse nor has unnecessary broken boundaries.

[0076] As a preferred embodiment of the above embodiments, the value at a specific quantile position is not less than half the standard lane width.

[0077] Specifically, the reason for this setting is that if the α value is set too small, the generated driving area boundary will be too close to the point cloud, resulting in very fragmented, tortuous, or broken conditions. This will fail to accurately reflect the overall continuous driving space of vehicles in real traffic. In actual driving scenarios, the instantaneous trajectory of a vehicle is affected by random factors such as turning, lane changing, acceleration, and deceleration. Some subtle fluctuations or short-term offsets will appear in the point cloud. If the α value is less than half the standard lane width, these tiny trajectory fluctuations will be excessively amplified by the algorithm, leading to severe distortion of the area boundary and loss of the ability to express the overall lane or road network level space. Setting the lower limit of the α value to no less than half the standard lane width is precisely to ensure that the generated boundary has reasonable smoothness and integrity, can accommodate reasonable offsets and random fluctuations generated by vehicles during normal driving, and maintains a shape and width that conforms to the actual road structure. For example, when the lane width is 3.5 meters, setting the α value to no less than 1.75 meters can ensure that the generated area covers at least one main lane, avoiding the driving area being incorrectly divided into multiple narrow and discontinuous segments due to overfitting.

[0078] As a preferred embodiment of the above, such as Figure 3 As shown, outlier filtering for spatiotemporal point clouds includes:

[0079] B10: Detect stationary vehicle point clouds in spatiotemporal point clouds whose speed is continuously below a set threshold, and identify related point clouds that have spatial continuity with the stationary vehicle point clouds;

[0080] B20: Isolation points that are discontinuously distributed and have a velocity exceeding a threshold are marked as outliers;

[0081] B30: Adjust the filter radius based on the spatial density distribution of outliers and remove all outliers within the filter radius.

[0082] Specifically, the outlier filtering process for spatiotemporal point clouds begins with velocity information. Point clouds whose speeds consistently fall below a set threshold over a continuous time period are detected, identifying point sets representing stationary vehicles. Generally, prolonged stationary states in traffic are due to traffic light waiting, roadside parking, or severe congestion. To more accurately distinguish point cloud clusters formed under these stationary conditions, their spatial continuity can be further analyzed. If multiple low-speed points maintain a relatively fixed spatial distribution pattern and are connected to surrounding trajectories, these spatially continuous points are considered as associated point clouds of stationary vehicles and retained for subsequent dynamic obstacle analysis. Simultaneously, to eliminate invalid points caused by detection errors, image noise, or sporadic events, all points whose speeds exceed the threshold but lack trajectory continuity or neighborhood support need to be identified. These points... Spatially, outliers typically exhibit isolated distributions, neither belonging to the continuous trajectories of normal vehicles nor forming stable static structures. A neighborhood-based search approach is preferred to determine their isolation. For example, a certain spatial distance threshold is set; if there are insufficient neighboring points within this range, the point is marked as an outlier and identified as an object requiring filtering. After initial identification, to further improve the adaptability and accuracy of filtering, the filtering radius needs to be dynamically adjusted based on the density distribution characteristics of outliers throughout the space. Specifically, with each outlier as the center, the fluctuation of point cloud density within a certain radius is analyzed, and the variance of the density is calculated. If the variance is significantly higher than the set standard deviation threshold, it indicates that the point distribution in this area is unstable and prone to false positives or drift points. All marked outliers are then uniformly removed to generate a purer and more stable spatiotemporal point cloud.

[0083] As a preferred embodiment of the above, such as Figure 4 As shown, adjusting the filtering radius based on the spatial density distribution of outliers includes:

[0084] C10: Establish a radial distance function centered on outliers and calculate the variance of point cloud density within a preset range;

[0085] C20: When the variance exceeds the standard deviation threshold, the filter radius is increased proportionally.

[0086] C30: When the variance value does not exceed the standard deviation threshold, the initial fixed radius value is used for filtering.

[0087] Specifically, firstly, a radial region is established centered on each identified outlier, with distance as the variable. Within this region, the distribution density of the point cloud at different distance intervals is statistically analyzed. Then, the variance of the point cloud density over the entire preset range is calculated. This variance reflects the degree of fluctuation in the point cloud distribution within this region, i.e., whether there are spatial characteristics of drastic density changes and structural instability around the outlier. If the calculated density variance exceeds a preset standard deviation threshold, it indicates that the point cloud distribution within the outlier's region is uneven, potentially containing interfering factors such as false detections, drift points, or trajectory remnants. In this case... To more comprehensively eliminate these potential outliers, it is preferable to increase the filtering radius proportionally. The increase can be dynamically set based on the degree of density fluctuation. For example, when the variance exceeds the threshold by 1 time, the radius can be increased to 1.5 times or 2 times the original value to ensure that outliers in a larger area are removed simultaneously. Conversely, if the variance value does not exceed the set standard deviation threshold, it indicates that the point cloud distribution in the area is relatively stable, and the outliers are more likely to be isolated outliers without a large interference range. In this case, it is preferable to use the initially set fixed radius value for local filtering, which can efficiently remove invalid points and avoid accidentally deleting adjacent valid trajectory data.

[0088] As a preferred embodiment of the above, such as Figure 5 As shown, step S1, obtaining the vehicle location coordinates in the traffic image and constructing a spatiotemporal point cloud, includes:

[0089] S11: Extract vehicle position coordinates from traffic images using object detection algorithms;

[0090] S12: Convert the extracted vehicle location coordinates into three-dimensional spatiotemporal coordinates, where two dimensions represent the geographic location and the third dimension represents the timestamp;

[0091] S13: Associate the spatiotemporal coordinates of the same vehicle within a continuous time period to form a motion trajectory, and aggregate the motion trajectory data of all vehicles to generate a spatiotemporal point cloud.

[0092] Specifically, firstly, continuous traffic image sequences are acquired using surveillance cameras deployed in road scenes. A deep learning-based object detection algorithm (such as YOLOv5, Faster R-CNN, or CenterNet) is preferably used to analyze each frame, automatically extracting the vehicle's position coordinates within the image. These coordinates are typically represented as two-dimensional pixel boxes or keypoints. This detection process is robust, consistently outputting vehicle detection results under varying lighting, weather, and traffic density conditions, ensuring high accuracy of the position information relied upon for subsequent processing. Next, the vehicle position coordinates extracted from each frame are transformed from pixel space to geographic space. Using the camera's calibration parameters (including intrinsic and extrinsic parameters), camera installation height, and pitch angle, a perspective transformation or homography mapping model is constructed to convert the image coordinates to the actual two-dimensional coordinates. Precise mapping of 3D planar coordinates is crucial. For example, the center point of a vehicle in an image can be mapped to ground coordinates in a road coordinate system measured in meters. Combined with the image frame's acquisition timestamp, each coordinate point is expanded into a 3D spatiotemporal coordinate point containing both spatial location and temporal information. The first two dimensions represent the vehicle's geographical location, and the third dimension represents the time of the vehicle's appearance. To further enhance the completeness and trajectory coherence of the spatiotemporal point cloud, it is necessary to correlate the position points of the same vehicle in consecutive frames. This can be achieved using multi-target tracking algorithms such as appearance feature matching, Kalman filter prediction, or optical flow to identify and track the continuous movement of the same vehicle over time, thus forming a complete vehicle trajectory. By aggregating all identified vehicle trajectories, a large set of high-density, multi-time-phase 3D points can be generated, forming a complete spatiotemporal point cloud. This point cloud not only preserves the actual spatial distribution of vehicles but also accurately expresses the dynamic evolution of vehicles over time, realistically reproducing the motion characteristics of traffic flow.

[0093] As a preferred embodiment of the above, such as Figure 6 As shown, in step S4, the α value is input into the Alpha Shape algorithm to generate the road network-level driving area, and the α value is applied independently to each point cloud cluster to generate the lane-level driving area, including:

[0094] S41: Input the global α value into the Alpha Shape algorithm to reconstruct the surface of the filtered complete spatiotemporal point cloud and generate a continuous driving area covering the entire road network;

[0095] S42: Based on the parallelism metric of vehicle motion vectors, the spatiotemporal point cloud is segmented into independent point cloud clusters;

[0096] S43: Input the same α value as the road network level for each point cloud cluster, and generate the lane-level driving area through the Alpha Shape algorithm.

[0097] Specifically, firstly, the global α value calculated in the previous step is applied to the Alpha Shape algorithm, and the complete spatiotemporal point cloud after outlier filtering is input for surface reconstruction. The purpose of this operation is to generate a continuous driving area covering the entire road network, reflecting the actual travel boundaries of all vehicles at different times and locations. By performing spatial envelope processing on the complete point cloud using the AlphaShape algorithm, the overall activity space contour of the vehicles can be effectively extracted, forming a macroscopic, continuous road network-level driving area, providing basic support for city-level traffic management and route planning. Based on the generated road network-level area, to further improve the refinement of the area extraction, the point cloud needs to be divided into multiple independent subsets representing different lanes, based on vehicle motion. Parallelism of vectors is used to cluster point clouds, grouping point clouds with similar motion directions and velocity change trends into the same cluster. The motion trajectories of vehicles within the same lane usually exhibit a high degree of parallelism. Therefore, by calculating the velocity direction vectors between continuous trajectory points and setting an angle threshold to measure their similarity, point cloud clusters of different lanes can be effectively separated. For example, in an intersection area containing multiple straight and left-turn lanes, the motion vectors of vehicles in different directions are clearly distinguishable. This method can clearly divide multiple point cloud clusters with consistent motion directions. Subsequently, the same α value as the generated road network-level region is input into each point cloud cluster, and the Alpha Shape algorithm is applied again to reconstruct the local boundaries, resulting in a refined lane-level driving region. This operation can extract precise contours reflecting the shape, width, and direction of individual lanes while maintaining overall boundary consistency, further improving the resolution and expressiveness of the region. For example, in a typical embodiment, vehicles in a four-lane main road travel stably according to their respective lane directions. The clustering process can generate four point cloud clusters with consistent motion and extract four clear and naturally defined lane-level driving regions. By using the above method, the region is generated hierarchically from macro to micro, which not only ensures the continuity and precision of the driving region but also provides high-quality spatial references for functions such as lane keeping and lane change decision-making in autonomous driving.

[0098] As a preferred embodiment of the above, such as Figure 7 As shown, in step S5, a dynamic obstacle zone is generated based on the filtered stationary vehicle position points. The generated road network-level region and lane-level driving region are integrated to output a multi-level driving region for vehicles, including:

[0099] S51: Generates a dynamic obstacle avoidance zone based on static obstacle point cloud. When it intersects with the original lane-level area space, it reconstructs and generates a continuous navigable lane-level channel in real time.

[0100] S52: Integrates the road network-level area, lane-level continuous navigable channel, and dynamic obstacle avoidance zone into a three-layer structure, outputting a multi-level driving area for vehicles.

[0101] Specifically, firstly, stationary vehicle location points retained during the outlier filtering process are extracted. These points typically represent vehicles stationary on the road due to traffic congestion, temporary parking, or unforeseen events. An obstacle point cloud is constructed based on these static points. Then, a buffer zone expansion method is preferably used to generate a dynamic obstacle avoidance zone. This involves setting a safe buffer radius around each stationary point to construct a space area that other vehicles need to detour through. This obstacle avoidance zone can be dynamically updated and adjusted in real-time according to traffic flow changes to reflect temporary obstructions that may occur on the road at any time. When the dynamic obstacle avoidance zone spatially overlaps or interferes with the previously extracted lane-level driving area, a continuous navigable lane-level channel is reconstructed in real-time based on the remaining passable space. Through spatial Boolean operations, the overlapping parts of the lane-level region with the obstacle avoidance zone are removed, and the unoccupied space area is retained as the actual passable lane area. If the original lane is interrupted due to a stationary obstacle, the boundaries of the adjacent lane area can be expanded or merged according to the point cloud trajectory direction and density to generate a spatially continuous and passable alternative path. On this basis, the road network-level region, the lane-level continuous navigable channel, and the dynamic obstacle avoidance zone are spatially integrated to construct a complete three-layer structure. This structure represents the overall passage space, the high-precision channel reflected by the actual driving trajectory of the vehicle, and the temporary detour area caused by dynamic and stationary objects, covering multi-level driving needs from macro-global to micro-avoidance.

[0102] Example 2

[0103] Based on the same inventive concept as the vehicle driving area generation method in the foregoing embodiments, the present invention also provides a vehicle driving area generation system, the system comprising:

[0104] The point cloud construction module acquires the vehicle location coordinates in traffic images and constructs a spatiotemporal point cloud.

[0105] The lane clustering module filters outliers from the spatiotemporal point cloud and performs lane clustering based on the filtered spatiotemporal point cloud to separate point cloud clusters that represent different lanes.

[0106] The α value calculation module calculates the α value based on the spatial density of spatiotemporal point clouds. The α value is the maximum radius of the circumcircle of the generated polygon as defined in the Alpha Shape algorithm.

[0107] The α value application module inputs the α value into the Alpha Shape algorithm to generate a road network-level driving area, and independently applies the α value to each point cloud cluster to generate a lane-level driving area;

[0108] The region output module generates dynamic obstacle zones based on the filtered stationary vehicle location points, integrates the generated road network-level regions and lane-level regions, and outputs multi-level driving regions for vehicles.

[0109] The generation system described above in this invention can effectively realize the vehicle driving area generation method, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0110] As a preferred embodiment of the above, the α value application module includes:

[0111] The continuous region generation unit inputs the global α value into the Alpha Shape algorithm to reconstruct the surface of the filtered complete spatiotemporal point cloud, generating a continuous driving region covering the entire road network;

[0112] The spatiotemporal point cloud segmentation unit, based on the parallelism metric of vehicle motion vectors, segments the spatiotemporal point cloud into independent point cloud clusters;

[0113] The lane-level region generation unit inputs the same α value as the road network-level unit to each point cloud cluster and generates lane-level driving regions using the AlphaShape algorithm.

[0114] Similarly, the above-mentioned optimization schemes for the system can also achieve the optimization effects corresponding to the methods in Embodiment 1, which will not be repeated here.

[0115] Although this application has been described in conjunction with specific features and embodiments, it will be apparent that various modifications and combinations can be made therein without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application.

[0116] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for generating a vehicle driving area, characterized in that, include: Obtain vehicle location coordinates from traffic images and construct a spatiotemporal point cloud; Outlier filtering is performed on the spatiotemporal point cloud, and lane clustering is performed based on the filtered spatiotemporal point cloud to separate point cloud clusters representing different lanes. The value of α is calculated based on the spatial density of the spatiotemporal point cloud. The value of α is the maximum radius of the circumcircle of the generated polygon as defined in the Alpha Shape algorithm. The α value is input into the Alpha Shape algorithm to generate a road network-level driving area, and the α value is applied independently to each point cloud cluster to generate a lane-level driving area; Dynamic obstacle zones are generated based on the filtered stationary vehicle location points. The generated road network-level areas and lane-level driving areas are integrated to output multi-level vehicle driving areas.

2. The method for generating a vehicle driving area according to claim 1, characterized in that, The calculation of the α value based on the spatial density of the spatiotemporal point cloud includes: The filtered spatiotemporal point cloud is divided into equal-sized grid units; Calculate the number of points in each of the grid cells, and determine the local α value based on the inverse relationship of the number of points. The local α values ​​are sorted, and the values ​​at specific quantile positions in the sort are selected as the global α values.

3. The method for generating a vehicle driving area according to claim 2, characterized in that, The value at the specific quantile position is not less than half the standard lane width.

4. The method for generating a vehicle driving area according to claim 1, characterized in that, Outlier filtering of the spatiotemporal point cloud includes: Detect stationary vehicle point clouds in the spatiotemporal point cloud whose velocity is continuously below a set threshold, and identify associated point clouds that have spatial continuity with the stationary vehicle point clouds; Isolated points that are discontinuously distributed and have a velocity exceeding a threshold are marked as outliers; The filtering radius is adjusted based on the spatial density distribution of the outliers, and all outliers within the filtering radius are removed.

5. The method for generating a vehicle driving area according to claim 4, characterized in that, Adjusting the filtering radius based on the spatial density distribution of the outliers includes: A radial distance function is established with the outlier point as the center, and the variance of the point cloud density within a preset range is calculated; When the variance value exceeds the standard deviation threshold, the filtering radius is increased proportionally. When the variance value does not exceed the standard deviation threshold, an initial fixed radius value is used for filtering.

6. The method for generating a vehicle driving area according to claim 1, characterized in that, Collect vehicle location coordinates from traffic images and construct a spatiotemporal point cloud, including: Vehicle position coordinates are extracted from the traffic image using an object detection algorithm; The extracted vehicle location coordinates are converted into three-dimensional spatiotemporal coordinates, where two dimensions represent the geographic spatial location and the third dimension represents the timestamp; The spatiotemporal coordinates of the same vehicle within a continuous time period are associated to form a motion trajectory, and the motion trajectory data of all vehicles are aggregated to generate the spatiotemporal point cloud.

7. The method for generating a vehicle driving area according to claim 1, characterized in that, The α value is input into the AlphaShape algorithm to generate a road network-level driving region, and the α value is independently applied to each point cloud cluster to generate a lane-level driving region, including: The global α value is input into the Alpha Shape algorithm to reconstruct the surface of the filtered complete spatiotemporal point cloud, generating a continuous driving area covering the entire road network; Based on the parallelism metric of vehicle motion vectors, spatiotemporal point clouds are segmented into independent point cloud clusters; Input the same α value as the road network level to each point cloud cluster, and generate lane-level driving areas using the Alpha Shape algorithm.

8. The method for generating a vehicle driving area according to claim 1, characterized in that, Dynamic obstacle zones are generated based on the filtered stationary vehicle location points. These are then integrated with the generated road network-level and lane-level driving zones to output a multi-level vehicle driving zone, including: Dynamic obstacle avoidance zones are generated based on static obstacle point clouds. When these zones intersect with the original lane-level area space, lane-level continuous navigable channels are reconstructed in real time. The road network-level area, lane-level continuous navigable channel, and dynamic obstacle avoidance zone are integrated into a three-layer structure to output the vehicle's multi-level driving area.

9. A vehicle driving area generation system, characterized in that, The system includes: The point cloud construction module acquires the vehicle location coordinates in traffic images and constructs a spatiotemporal point cloud. The lane clustering module filters outliers from the spatiotemporal point cloud and performs lane clustering based on the filtered spatiotemporal point cloud to separate point cloud clusters that represent different lanes. The α value calculation module calculates the α value based on the spatial density of the spatiotemporal point cloud. The α value is the maximum radius of the circumcircle of the generated polygon as defined in the Alpha Shape algorithm. The α value application module inputs the α value into the Alpha Shape algorithm to generate a road network-level driving area, and independently applies the α value to each point cloud cluster to generate a lane-level driving area; The region output module generates a dynamic obstacle zone based on the filtered stationary vehicle location points, integrates the generated road network-level region and lane-level region, and outputs a multi-level driving region for vehicles.

10. The vehicle driving area generation system according to claim 9, characterized in that, The α value application module includes: The continuous region generation unit inputs the global α value into the Alpha Shape algorithm to perform surface reconstruction on the filtered complete spatiotemporal point cloud, generating a continuous driving region covering the entire road network; The spatiotemporal point cloud segmentation unit, based on the parallelism metric of vehicle motion vectors, segments the spatiotemporal point cloud into independent point cloud clusters; The lane-level region generation unit inputs the same α value as the road network-level value to each of the point cloud clusters and generates lane-level driving regions using the AlphaShape algorithm.

Images