Road end point cloud rasterization method based on mask and white list area extraction
Through the rasterization method of road endpoint cloud extraction based on mask and whitelist area extraction, the existing road endpoint cloud detection methods are solved, and efficient and accurate perception of road end devices is achieved, adapting to dynamic traffic environments, reducing resource waste and boundary errors, and supporting parallel processing of multiple obstacles.
Patent Information
- Application Number
- CN202510564062.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-08
AI Technical Summary
The existing road endpoint cloud detection methods have shortcomings in detection accuracy and real-time performance, especially when fusion of vehicle-intensive areas and multi-sensors, it is difficult to balance detection accuracy and efficiency, and the existing methods handle at the point cloud level lead to resource waste and boundary errors.
Through the rasterization method of road-end point cloud extracted based on mask and whitelist area, the mapping relationship between multiple road-end devices and real scenes is established, the mapping relationship between point cloud and world coordinate system is constructed, and the RTK coordinates of temporary obstacles and whitelist areas are obtained in the real scene, and the regions of interest are directly divided under the physical world coordinate system, obstacles are removed and whitelist points are restored, so as to achieve efficient data processing.
It significantly improves the accuracy and real-timeness of road-end equipment detection, reduces the complexity of large-scale deployment, ensures the boundary accuracy of sparse point cloud areas, supports parallel processing of multiple obstacles, balances the accuracy and real-timeness of the algorithm, and meets the efficient and accurate perception needs of intelligent transportation systems.
Smart Images
Figure CN120451470A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of road endpoint cloud target detection, and in particular to a road endpoint cloud rasterization method based on mask and whitelist area extraction. Background Art
[0002] With the rapid development of intelligent connected and autonomous driving technologies, vehicle-infrastructure cooperative systems hold significant potential for road traffic monitoring, roadside anomaly detection, and improved traffic management optimization. Roadside perception fusion technology, a key component of vehicle-infrastructure cooperative systems, deeply integrates vehicle, road infrastructure, and cloud data to acquire and analyze traffic environment information in real time, enabling comprehensive monitoring of traffic flow, driving status, and potential risks.
[0003] During the road-end point cloud perception process, effective preprocessing of the original point cloud is required to ensure perception accuracy and real-time performance. However, the traditional rasterization method has a single adjustment mode during road-end detection and cannot achieve the segmentation of important areas and non-detection areas. Therefore, it is difficult to balance detection accuracy and real-time performance, resulting in a waste of hardware resources.
[0004] Furthermore, existing methods for detecting roadside point cloud targets generally employ two approaches: one is to directly detect targets at the point cloud level and then filter the region of interest; the other is to first filter the region of interest and then detect targets within it. Both methods ultimately rely on radar coordinates to map detection results to the real-world scene. Their core processing steps are concentrated at the point cloud level, and grid filtering is commonly used for point cloud preprocessing. However, in areas with dense vehicle traffic, these methods often need to discard a large amount of valid point cloud data to meet real-time requirements, severely impacting detection accuracy. More importantly, when multi-sensor fusion detection is required, since all processing is performed at the point cloud level, the delineation of regions of interest in real-world scenarios becomes extremely complex. This not only significantly increases the time cost of algorithm deployment but also severely restricts the performance optimization and iterative upgrades of roadside perception systems, making it difficult for existing methods to meet the urgent needs of intelligent transportation systems for efficient and accurate perception. Summary of the Invention
[0005] The present invention aims to provide a road end point cloud rasterization method based on mask and whitelist area extraction to solve the problems of inaccurate and insufficient real-time detection of existing road end equipment.
[0006] To achieve the above objectives, the present invention adopts the following technical solutions: a road endpoint cloud rasterization method based on mask and whitelist area extraction, which performs multiple types of processing on various areas where roadside vehicles travel, including:
[0007] Step 1: Establish a mapping relationship between multiple road-end devices and real scenes, and construct a mapping relationship between point clouds and the world coordinate system;
[0008] Step 2: rasterize the point cloud data in the world coordinate system, establish a mapping relationship between the point cloud data and the grid, and construct a grid coordinate system;
[0009] Step 3: Obtain the RTK coordinates of temporary obstacles and whitelist areas in the real scene and convert them into a grid coordinate system;
[0010] Step 4: Process the point cloud data at the obtained coordinates, remove obstacles and restore the whitelist points to obtain road-side perception data.
[0011] The principles and advantages of this solution are:
[0012] Existing roadside detection methods typically use global point cloud filtering to balance accuracy and efficiency, but this struggles to address the local variations required in complex traffic scenarios. In particular, in real-world roadside scenarios, sufficient point clouds are required to characterize vehicles in both densely populated and sparsely populated areas, particularly queuing areas. However, existing regional gridding methods primarily focus on global point cloud filtering, lacking the ability to process localized point clouds. This rigid processing model results in both wasted resources in sparse areas and insufficient characterization in key regions, making existing point cloud regionalization less adaptable.
[0013] Secondly, in large-scale roadside deployments, the existing method is to select regions at the point cloud level. However, this method requires individual processing for each sensor. Furthermore, the point cloud data cannot avoid boundary errors caused by missing point cloud data when dividing regions of interest, resulting in large errors in the existing division method. Furthermore, during roadside obstacle detection, obstacles added temporarily at intersections are often temporarily removed. However, when removing obstacles at the point cloud level, this cannot be effectively linked to the actual scene. Instead, it is more of a rough range screening at the point cloud data level, resulting in the loss of some point cloud data when vehicles enter that area.
[0014] When integrating multiple sensors, existing methods require dividing each type of sensor into separate areas. However, since each type of sensor acquires different types of data, the division methods are also diverse, which increases the complexity when deployed on a large scale on the roadside.
[0015] This solution fundamentally addresses the limitations of existing roadside detection methods in region demarcation and point cloud processing through a unified mapping of the world coordinate system. By completing global coordinate system conversion during the data preprocessing phase, multi-lidar data can be fused and processed under a unified spatial reference, thus avoiding boundary errors caused by individual sensor differences or point cloud sparsity. Furthermore, by extracting intersection boundary feature points from real-world scenarios and directly demarcating regions of interest in the physical world coordinate system, this solution not only reduces the complexity of large-scale deployment but also ensures boundary accuracy in sparse point cloud areas, significantly improving the system's standardization and scalability.
[0016] Secondly, this solution completes obstacle suppression during the rasterization phase, avoiding redundant calculations while supporting parallel processing of multiple obstacles, significantly improving algorithm efficiency. This not only solves the point cloud discontinuity problem during temporary obstacle removal, but also ensures scene semantic consistency, enabling more stable adaptation to dynamic traffic environments. Furthermore, this solution achieves differentiated processing for different regions through whitelisting and boundary fitting. By establishing a precise mapping relationship between the point cloud and the grid, and restoring the original point cloud using boundary feature points, this approach reduces data volume while retaining necessary detail information. This effectively balances the accuracy and real-time nature of the algorithm, ensuring the efficient and precise perception requirements of intelligent transportation. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 The figure is a flow chart of the method for rasterizing the road endpoint cloud based on mask and whitelist area extraction according to the present invention. DETAILED DESCRIPTION
[0018] The following is further described in detail through specific implementation methods:
[0019] Example 1
[0020] The road end point cloud rasterization method based on mask and whitelist area extraction in this embodiment is based on the mask area extraction method proposed by perspective transformation, and combined with the area whitelist algorithm, it performs multiple types of processing on various areas where road end vehicles travel. This improves the accuracy and real-time performance of road end equipment detection and provides effective data support for road end perception. In this embodiment, as shown in the attached Figure 1 As shown, including:
[0021] S1, establish the mapping relationship between multiple road-end devices and real scenes, and construct the mapping relationship between point cloud and world coordinate system.
[0022] In this embodiment, the test environment is first selected and set up. During the intelligent vehicle network testing process, environmental scenario preparation is required. A good test road can clearly reflect the effects and problems of algorithm deployment. Ideally, the test road should cover multiple road environments, including but not limited to urban intersections with complex traffic flows and highway through lanes. Such roads can ensure the acquisition of more diverse road condition information. Therefore, after selecting the corresponding test scenario, the sensing equipment is installed. In this embodiment, the safety of the sensing equipment must ensure that it can cover the entire detection area of the selected road.
[0023] After preparing the test environment, select at least 10 known spatial points from the selected test intersection (i.e., the desired real-world scenario) as reference points. Ensure that the reference points are distributed throughout the test area to avoid large global errors caused by local measurement errors. This ensures that the original calibration data has accurate position coordinates, thereby reducing coordinate mapping errors. The coordinate values of the selected reference points are then recorded in the world coordinate system and in the roadside equipment.
[0024] Based on the selected reference point, a mapping relationship matrix between the world coordinate system and the road-end equipment data coordinates is established through perspective transformation, that is, multiple sensor data (point cloud data) are unified into the world coordinate system.
[0025] In this embodiment, the point cloud data is mapped to the real-world coordinate system to construct a mapping relationship between the point cloud and the world coordinate system. The perspective transformation matrix H is a 3x3 matrix that maps the point cloud coordinates (x, y, z) to the real-world coordinate system (X, Y, Z). Its expression is:
[0026]
[0027] In practice, H is calculated from at least 10 known corresponding points using the least squares method or SVD decomposition. In the real-world coordinate system, since multiple radar point clouds need to be covered, the Z-axis value is usually taken as the original value.
[0028] At the same time, in order to further improve the accuracy of perspective transformation, in this embodiment, the Levenberg-Marquardt algorithm is used for matrix optimization, and the perspective transformation matrix H is adjusted by combining the gradient descent method and the Newton method. Its expression is:
[0029]
[0030] Where, J k is the Jacobian matrix; δ is the adjustment factor; r k is the residual; I is the identity matrix.
[0031] S2, rasterizes the point cloud data in the world coordinate system, establishes a mapping relationship between the point cloud data and the grid, and constructs a grid coordinate system.
[0032] In this embodiment, the world coordinate system is the UTM coordinate system, which is mainly the Mercator projection coordinate system, and its expression is
[0033]
[0034] Where R is the radius of the Earth, λ is the longitude, and φ is the latitude.
[0035] At the same time, since the coordinate values of the point cloud data after being projected into the real coordinate system are very large, it is also necessary to centralize the point cloud so that the center of the grid point cloud coordinates is the origin. The expression is:
[0036]
[0037] In the formula, (x u, y u ) is the UTM coordinate system point; (x 0, y0) is the center point, (x H, y H ) are the coordinate points for calculating the H matrix.
[0038] In order to establish a mapping relationship between the point cloud coordinates and the real world and effectively format the point cloud, in this embodiment, it is necessary to rasterize the point cloud and construct a grid coordinate system. Rasterization is to map the point cloud after perspective transformation to the grid coordinate system according to the distance and position relationship of the point cloud, and perform filtering according to the set grid size. The expression of rasterization is:
[0039]
[0040] Among them, (x g ,y g ) is the coordinate point of the grid coordinate system; r x 、r y Respectively represent the size of the horizontal and vertical coordinates; s is the size of each grid.
[0041] S3, obtains the RTK coordinates of temporary obstacles and whitelist areas in the real scene and converts them into the grid coordinate system.
[0042] In this embodiment, in order to effectively cover the intersection and ensure the real-time operation of the algorithm, the boundary area of the intersection must be extracted and the unnecessary detection area must be effectively removed, thereby reducing the data input when the algorithm is running.
[0043] Therefore, in this embodiment, the boundary area of the intersection is first determined to obtain a valid target area. In actual operation, the intersection boundary feature points under real conditions are obtained through RTK equipment as mask points. The mask points are converted into grid coordinate points centered in the UTM coordinate system under the established mapping relationship. In this way, mask boundary description points that conform to the real relationship are constructed. The mask network is constructed in the form of a point cloud grid with the same amount of points. The area surrounded by the mask boundary points is obtained by polygonal area filling, thereby extracting valid grid coordinates and obtaining the target area.
[0044] In this embodiment, polygonal area filling includes the following process:
[0045] Determine the mask boundary: Scan row by row along the x-axis, and calculate the minimum and maximum boundary values and the scanning intersection points on the y-axis and the polygon area.
[0046] Determine the pixels to be filled: Sort the obtained intersection list by x-coordinate to determine the pixels to be filled.
[0047] Pixel filling: Based on the sorted intersection points, fill the pixels between the intersection points in sequence according to the set filling data.
[0048] Complete region acquisition: The target region is gradually constructed until all scan lines are processed and the entire region is filled, thus obtaining the target region. This ensures that in intersection scenarios, the effective intersection area is extracted, the non-detection area is shielded, the algorithm operation efficiency is optimized, and unnecessary data processing is reduced.
[0049] After the target area is determined, temporary obstacles and whitelist areas are obtained within the target area. In this embodiment, object coordinates are obtained from the real road conditions, and the whitelist areas are distinguished and marked.
[0050] In complex traffic scenarios, such as traffic congestion, motorcycles / bicycles and cars driving side by side, etc., traditional detection algorithms may merge adjacent vehicles into large trucks or miss some targets because the target objects are too close or the point cloud features are blurred, such as edge extension, particle offset, etc., resulting in misidentification and affecting data accuracy. In this embodiment, the whitelist area is to define the precise independent areas of multiple identified objects, clarify the actual location of the whitelist area, ensure that the accuracy of the original data can be retained after subsequent rasterization processing, and avoid feature confusion or mismerging. In this way, the real object distribution is restored, the detection accuracy is improved, and the amount of calculation is reduced during rasterization, repeated detection and redundant calculations are reduced. By marking high-frequency detection in known areas, it can avoid misjudgment in actual operations, save computing power, and optimize efficiency and reliability.
[0051] In this embodiment, the whitelist area is a marked area demarcated by identifying multiple local areas where dynamic objects such as vehicles, pedestrians, motorcycles, etc. are located in the target area. Specifically, the real whitelist area enclosing points at the road end are obtained through the RTK equipment or the UTM visual coordinate interface, and the RTK coordinates or the selected UTM coordinate points are converted to the grid according to the constructed mapping relationship. Among them, the area surrounded by the enclosing points is called the "whitelist area". In subsequent rasterization processing (such as map segmentation, obstacle analysis), even if the point cloud data is rasterized, the features can still be supplemented and restored through the marked whitelist area, thereby ensuring the integrity and accuracy of the original features. And reduce the amount of calculation during rasterization, and improve real-time performance.
[0052] In this embodiment, the obstacle is an area formed by temporarily placed objects that affect traffic or need to be avoided. In order to better remove obstacles at the intersection, it is necessary to obtain the coordinate points of the obstacle from the actual road conditions.
[0053] In this embodiment, obstacle point coordinates are extracted using the same coordinate extraction method as the mask boundary. This involves using the RTK device to select at least four corner points near the detected obstacle to obtain the coordinates of the obstacle point under real-world conditions. Based on the grid resolution and origin, the grid number of the obstacle is calculated to form the obstacle grid index. Furthermore, if the obstacle has volume, such as a display stand, the coordinates of the four corner points of its circumscribed rectangle are converted into a grid index, marking the entire occupied area to quickly determine whether the path is occupied by the obstacle. This method allows for the rapid addition or removal of obstacles in real time, directly reflecting environmental changes by judging the indexed grids. This optimizes computational efficiency while improving the real-time and accuracy of dynamic judgments.
[0054] S4, processes the point cloud data under the obtained coordinates, removes obstacles and restores the whitelist points to obtain road-end perception data.
[0055] In existing technologies, obstacle removal typically involves filtering perception results. This approach requires target traversal after each detection, increasing algorithm time and compromising real-time performance. This solution, however, removes obstacles by modifying background grid points using obstacle corner points acquired by RTK during the initialization phase.
[0056] Modifying the background grid points includes the following steps:
[0057] Convex hull operation of obstacle corner points: rasterize the obstacle points and obtain the minimum set of contour points through the corner points.
[0058] Obstacle area filling: Use the contour points as the boundary to fill the enclosed area in the grid and obtain the coordinates of the grid points in the obstacle area.
[0059] Background point expansion interpolation: Since the point cloud is sparse, the rasterized points are also sparse, and the background points need to be expanded and interpolated to increase the density of the background points. Therefore, the rectangular expansion method is used to expand and interpolate the background points, which is expressed as
[0060]
[0061] Among them, B is the rectangular structure element; A is the target to be expanded; Z is the coordinate of the expansion point.
[0062] Background point modification: Based on the grid points in the obstacle area, the background points that belong to the obstacle are modified to increase the Z value of the background points in the obstacle area. Since background points are removed by subtraction, increasing the value of the background points in the obstacle area can effectively remove the area before the detection phase, thereby reducing the runtime of the detection task.
[0063] In the process of point cloud rasterization, rasterization is to downsample the point cloud. However, in areas such as vehicle queues and sparse point cloud areas, downsampling will eliminate the inherent features of the point cloud, and it will inevitably lose some of the original features, resulting in blurred object contours and distorted geometric features, which in turn affects detection accuracy. Therefore, this solution innovatively introduces a whitelist area mapping mechanism. On the basis of point cloud rasterization, a raster mapping relationship is constructed, and the corresponding original point cloud data is directly reversely located based on the enclosing points of the whitelist area and restored. Compared with direct operations at the traditional point cloud level, this method uses the real scene as the coordinate basis, re-extracts the original point cloud information corresponding to the whitelist area, and realizes the restoration of the whitelist area, thereby ensuring the timeliness of data downsampling while retaining sufficient feature information related to detection, accurately restoring the key features lost during the downsampling process, ensuring that the geometric features of the key areas are not excessively damaged, ensuring the accuracy and integrity of the data, and ultimately achieving a balance between efficiency and accuracy.
[0064] In this embodiment, the whitelist area restoration process is as follows:
[0065] Sampling area points: Use RTK equipment to obtain the coordinates of the whitelist area in the real scene, and convert the real coordinates into the grid coordinate system through perspective transformation and rasterization to determine the position of the whitelist in the grid map.
[0066] Grid whitelist area construction: Construct the global whitelist grid coordinates in an area-filling manner, generate a closed area, and obtain the whitelist area grid index.
[0067] Point cloud mapping relationship: During the point cloud rasterization process, a corresponding mapping table between point cloud and raster is constructed, and a mapping relationship is established between each point cloud index and raster coordinate, so as to facilitate the rapid positioning of the original point cloud through the raster index.
[0068] Whitelist point restoration: Based on the whitelist raster index and raster mapping relationship, the corresponding original point cloud is found from the mapping table. Key points are filtered using a quantitative threshold method to restore the point cloud in the whitelist area, thereby recovering the downsampled features. The restoration level is controlled to restore the detailed features of the whitelist area while eliminating redundant points. This ensures filtering of the original point cloud, reducing the data volume of the original point cloud while retaining the original point cloud in areas with fewer features, effectively balancing the accuracy and real-time performance of the algorithm.
[0069] Existing roadside detection methods typically perform some degree of filtering on the roadside point cloud to reduce time costs while ensuring accuracy. However, in densely populated areas or on busy roads, uniform grid division weakens the point cloud's ability to represent key targets. The lack of adaptive local processing prevents dynamic resolution adjustment based on the scenario, severely limiting detection accuracy.
[0070] At the same time, existing methods rely on point cloud data itself for region segmentation. This fails to achieve standardized configuration across individual sensor variations and is susceptible to boundary drift due to missing points. When temporary obstacles appear at intersections, coarse-grained filtering based on point cloud ranges disrupts scene semantic consistency, causing point cloud discontinuities when vehicles pass through in real-world environments. The inherent flaws of this data-driven segmentation approach result in a lack of spatial cognition robustness in dynamic traffic scenarios, significantly increasing on-site commissioning and maintenance costs.
[0071] Furthermore, multi-sensor collaboration scenarios further expose the limitations of existing approaches. The diverse data types of heterogeneous sensors force each sensing unit to adopt independent regional division criteria. This fragmented processing model not only significantly increases integration complexity but also leads to cumulative errors in the spatiotemporal alignment of multi-source data. Existing technologies struggle to establish a unified spatial representation framework across sensors. Consequently, the high-precision, real-time perception capabilities urgently needed by intelligent transportation systems are constrained by underlying architectural flaws, making them unable to meet the stringent requirements for continuous, full-domain perception required by emerging applications such as vehicle-infrastructure collaboration.
[0072] This solution breaks through the inherent thinking mode of traditional road-side perception systems in coordinate system conversion and area division. Unlike conventional methods that perform post-processing at the sensor's local coordinate system or point cloud data level, this solution creatively achieves global spatial unification of multi-source lidar data during the data pre-processing stage. This fundamental architectural innovation enables all subsequent processing to be based on an accurate world coordinate system. By directly mapping the physical boundary features of the real scene to the digital space, rather than relying on easily missing point cloud features for area division, this solution cleverly circumvents the boundary drift problem caused by the sparsity of point clouds in traditional methods. This direct mapping relationship from the real physical world to the digital space constitutes the most groundbreaking technical feature of this solution.
[0073] Secondly, this solution takes a unique approach to obstacle handling. It pre-identifies the spatial characteristics of temporary obstacles in the world coordinate system. Through innovative dynamic threshold adjustment of background points, it achieves intelligent obstacle suppression during the initial stage of point cloud rasterization. This not only significantly reduces computational overhead but, more importantly, enables parallel real-time processing of multiple obstacles. This significant improvement in processing efficiency represents an order of magnitude improvement, enabling it to easily handle complex and ever-changing real-world traffic scenarios.
[0074] At the same time, this solution integrates adaptive processing methods of whitelist and boundary feature preservation. By constructing a bidirectional precise mapping relationship between point cloud and raster, and combining it with an innovative boundary fitting algorithm, it realizes intelligent identification and protection of key feature areas. It can dynamically adjust the processing method according to regional characteristics, while ensuring real-time performance, and still maintain high-precision perception of key areas such as sparse point clouds.
[0075] The above is only an embodiment of the present invention, and the common knowledge such as the specific technical solutions and / or characteristics in the solution are not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the description can be used to interpret the content of the claims.
Claims
1. The road endpoint cloud rasterization method based on mask and whitelist area extraction is characterized by: Perform various types of processing on various areas where vehicles travel on the road, including: Step 1: Establish a mapping relationship between multiple road-end devices and real scenes, and construct a mapping relationship between point clouds and the world coordinate system; Step 2: rasterize the point cloud data in the world coordinate system, establish a mapping relationship between the point cloud data and the grid, and construct a grid coordinate system; Step 3: Obtain the RTK coordinates of temporary obstacles and whitelist areas in the real scene and convert them into a grid coordinate system; Step 4: Process the point cloud data at the obtained coordinates, remove obstacles and restore the whitelist points to obtain road-side perception data.
2. The method for rasterizing a road endpoint cloud based on mask and whitelist area extraction according to claim 1, characterized in that: In step 1, at least 10 known reference points are selected from the real scene, and the reference points must cover the entire test area; the coordinate values of the reference points in the world coordinate system and the roadside equipment are recorded respectively.
3. The method for rasterizing road endpoint clouds based on mask and whitelist area extraction according to claim 2, characterized in that: Based on the selected reference points, a mapping relationship matrix between the world coordinate system and the road-end equipment data coordinates is established through perspective transformation, and multiple sensor data are unified into the world coordinate system.
4. The method for rasterizing a road endpoint cloud based on mask and whitelist area extraction according to claim 3, characterized in that: In step 2, the world coordinate system is a Mercator projection coordinate system; The rasterization is to map the point cloud after perspective transformation into a grid coordinate system according to the distance and position relationship of the point cloud, and perform filtering according to the set grid size. The expression of the rasterization is: Among them, (xg, yg) is the coordinate point of the grid coordinate system; rx, r y Respectively represent the size of the horizontal and vertical coordinates; s is the size of each grid.
5. The method for rasterizing road endpoint clouds based on mask and whitelist area extraction according to claim 1, characterized in that: In step 3, it also includes determining the boundary area of the intersection, forming a target area, and obtaining temporary obstacles and whitelist areas in the target area; the target area obtains the intersection boundary feature points under real conditions through the RTK equipment, as mask points, the mask points are converted into grid coordinate system points, and mask boundary description points that conform to the real relationship are constructed; and a mask network is constructed with the same amount of point cloud grids, and polygonal area filling is used to obtain the area surrounded by the mask boundary description points, and valid grid coordinates are extracted.
6. The method for rasterizing road endpoint clouds based on mask and whitelist area extraction according to claim 5, characterized in that: The polygon area filling includes the following process: Determine the mask boundary: Scan row by row with the x-axis as the scanning direction, and calculate the minimum and maximum boundary values and scanning intersection points on the y-axis and the polygon area; Determine the pixels to be filled: Sort the obtained intersection list by x-coordinate to determine the pixels to be filled; Pixel filling: Based on the sorted intersection points, fill the pixels between the intersection points in sequence according to the set filling data; Complete area acquisition: gradually build the target area until all scan lines are processed and the entire area is filled.
7. The method for rasterizing a road endpoint cloud based on mask and whitelist area extraction according to claim 1, characterized in that: The whitelist area is a marked area defined by identifying the local area where dynamic objects are located; the obstacle is an area formed by temporarily placed objects that affect passage or need to be avoided; the obstacle is marked by selecting at least four corner points near the detected obstacle using an RTK device, obtaining the coordinates of the obstacle points under real conditions, and converting them into an obstacle grid index.
8. The method for rasterizing a road endpoint cloud based on mask and whitelist area extraction according to claim 1, characterized in that: In step 4, during the operation phase of initializing background points, obstacles are removed by modifying background grid points using obstacle corner points.
9. The method for rasterizing road endpoint clouds based on mask and whitelist area extraction according to claim 1, characterized in that: In step 4, the area whitelist enclosing points are obtained in the real scene, and the whitelist area is restored based on the real scene coordinates.
10. The method for rasterizing road endpoint clouds based on mask and whitelist area extraction according to claim 8, characterized in that: Modifying the background grid points involves the following steps: Convex hull operation of obstacle corner points: rasterize the obstacle points and obtain the minimum set of contour points through the corner points; Obstacle area filling: Using the contour points as the boundary, fill the enclosed area in the grid and obtain the coordinates of the grid points in the obstacle area; Background point expansion interpolation: The background points are expanded and interpolated using the rectangular expansion method, which is expressed as Among them, B is a rectangular structural element; A is the target to be expanded; Z is the coordinate of the expansion point; Background point modification: Increase the Z value of the background point in the obstacle area based on the grid points in the obstacle area.