Highway boundary extraction method based on vehicle-mounted laser point cloud

Abstract

The application discloses a highway boundary extraction method based on a vehicle-mounted laser point cloud, and comprises the following steps: S1, collecting point cloud data of a road and ground objects on both sides of the road, and pre-processing the point cloud data; S2, acquiring control points corresponding to the point cloud data, classifying the control points to obtain type value point data, then performing three times Hemite interpolation processing on the type value point data, and finally distributing the elevation difference between the interpolated type value point data and the point cloud data to the point cloud data to perform elevation correction processing; and S3, firstly, extracting road point cloud data in the point cloud data subjected to the elevation correction processing by using a DBSCAN clustering algorithm in combination with density and reflection intensity filtering; then, extracting boundary points of the extracted highway point cloud data by using an Alpha Shape edge detection algorithm; and finally, fitting the highway boundary points by using a RANSC straight line fitting method to obtain a complete highway boundary.

Description

Technical Field

[0001] This invention relates to the field of highway boundary extraction based on vehicle-mounted laser point clouds. Background Technology

[0002] Traditional methods for highway boundary extraction rely on manual on-site surveys or 2D map data, which are not only time-consuming and labor-intensive but also inefficient. With the development of Mobile LiDAR technology, automatic highway boundary extraction has become possible. Mobile LiDAR systems can rapidly collect 3D point cloud data of the road surface and its surrounding environment while in motion, providing an efficient and accurate data acquisition method.

[0003] 3D laser measurement technology is a non-contact, proactive, and rapid method for acquiring dense 3D point clouds of an object's surface. It can directly acquire high-precision 3D point clouds from a distance, including attributes such as 3D coordinates, reflection intensity, and optical texture. These point clouds exhibit a massive and non-uniform spatial distribution. Notably, 3D laser measurement technology is unaffected by weather conditions and possesses a certain degree of penetration. Therefore, in the densely vegetated and complex terrain of southwestern mountainous areas, 3D laser measurement technology has become an important surveying and mapping tool, effectively supporting highway reconstruction and expansion projects.

[0004] In recent years, 3D laser mapping technology has been gradually applied to highway construction surveying and design. However, in most highway surveys, 3D laser mapping is limited to acquiring high-precision terrain information, and there is limited exploration of deeper aspects of 3D laser mapping data during highway expansion, such as accurate extraction of highway boundary lines and digitization of highway assets. Furthermore, laser mapping technology has not yet formed a comprehensive laser application technology system covering the entire lifecycle in highway reconstruction and expansion projects. At the same time, the in-depth application of 3D laser data in the highway engineering field lacks technological exploration. In highways, except for bridge and tunnel sections, the road edges are generally flat and lack clear boundaries, unlike urban roads which have distinct road edges. Therefore, highway point cloud data often lacks clearly defined road edge features. In contrast, general road laser point clouds may focus more on road geometry and different features, including road edges. Because highways lack clearly defined road edge features, the methods for highway boundary extraction differ from those for urban roads. Therefore, research on highway boundary extraction methods based on vehicle-mounted laser point clouds not only has significant practical application value for highway design, construction, management, and maintenance, but also has profound implications for promoting the development of intelligent transportation systems and protecting the environment.

[0005] Although 3D laser mapping technology has been applied in highway reconstruction and expansion projects, primarily focusing on providing high-precision ground elevation data, its application still faces some limitations. For example, in optimizing the accuracy of vehicle-mounted laser point clouds, research mainly targets specific constraints and environmental feature points, but these conditions may be difficult to meet in real-world scenarios, thus requiring more widely applicable and robust methods. Regarding laser point cloud road boundary extraction, while various methods based on image, point cloud, and data fusion exist, it is still necessary to enhance adaptability and extraction accuracy in complex environments, especially in road scenes with complex lighting or structures. Overall, current research still needs to further refine technologies and methods for practical application scenarios to improve the efficiency and accuracy of highway reconstruction and expansion projects. Summary of the Invention

[0006] The purpose of this invention is to provide a method for extracting highway boundaries based on vehicle-mounted laser point clouds, in order to solve the problems of poor adaptability and extraction accuracy of existing 3D laser mapping technology in complex environments.

[0007] To address the aforementioned technical problems, this invention provides a method for highway boundary extraction based on vehicle-mounted laser point clouds, comprising:

[0008] S1: Collect point cloud data of roads and features on both sides of the roads, and preprocess the point cloud data;

[0009] S2: Obtain the control points corresponding to the point cloud data, classify the control points to obtain the shape point data; then perform three Hemite interpolation processes on the shape point data; finally, use IDW to allocate the elevation difference between the interpolated shape point data and the point cloud data, and perform elevation correction processing on the point cloud data.

[0010] S3: First, the DBSCAN clustering algorithm combined with density and reflection intensity filtering is used to extract road point cloud data from the point cloud data after elevation correction; then, the AlphaShape edge detection algorithm is used to extract boundary points from the extracted highway point cloud data; finally, the RANSC line fitting method is used to fit the highway boundary points to obtain the complete highway boundary.

[0011] Furthermore, in step S1, the preprocessing of the point cloud data includes:

[0012] Construct a KD tree-shaped point cloud map based on the collected point cloud data;

[0013] Denoising the point cloud data;

[0014] The denoised point cloud data is then simplified.

[0015] Furthermore, the point cloud data denoising process includes sequentially applying Gaussian filtering, surface fitting filtering, normal vector filtering, and morphological filtering to denoise the point cloud data.

[0016] Furthermore, in step S2, the three Hemite interpolation processes performed on the type value point data include:

[0017] The parameterization of the model points is performed using the standardized cumulative chord length method to determine the nodal function values;

[0018] Shear loss is calculated using Bézier curves to determine nodal derivative values;

[0019] Select a set of interpolation points, where each interpolation point includes a nodal function value and a nodal derivative value;

[0020] By constructing interpolation basis functions and then linearly combining the interpolation basis functions with the corresponding function values ​​and derivative values, a cubic polynomial can be obtained.

[0021] The curve or surface for interpolation between interpolation points is obtained by calculating a cubic polynomial.

[0022] Furthermore, step S2 also includes longitudinal and lateral verification of the point cloud data before and after elevation correction processing.

[0023] Further, in step S3, the road point cloud data extracted from the elevation-corrected point cloud data using the DBSCAN clustering algorithm combined with density and reflection intensity filtering includes:

[0024] The DBSCAN clustering algorithm is used to distinguish points on the road surface from points on other non-road surfaces;

[0025] Points on the road surface are distinguished from points on other non-road surfaces based on the different point cloud densities;

[0026] Points on the road surface are distinguished from points on other non-road surfaces based on the different reflectivities of the point cloud.

[0027] Furthermore, in step S3, the boundary point extraction of the extracted highway point cloud data using the AlphaShape edge detection algorithm includes:

[0028] Perform Delaunay triangulation on the input point set to generate a triangulation that connects all input points and satisfies the Delaunay condition;

[0029] Based on the constructed Delaunay triangulation, each triangle or edge is evaluated, its circumcircle radius is calculated, and it is compared with a preset α value. If the circumcircle radius of a triangle or edge is less than or equal to the α value, the triangle or edge is considered to be part of the target shape; if the radius is greater than the α value, the triangle or edge is considered not to belong to the target shape and must be discarded.

[0030] Through the above filtering steps, the remaining triangles or edges constitute the target shape that meets the Alpha Shapes condition; based on this, the edges or boundaries of the shape are extracted to obtain the outline of the target object.

[0031] Furthermore, in step S3, the RANSC straight line fitting method is used to fit the highway boundary points, resulting in the complete highway boundary including:

[0032] A subset of point cloud data is selected by random sampling, a model is fitted on this subset, and the model is compared with all point cloud data. Based on a pre-set threshold, it is determined which data points fit the model, thus obtaining a more reliable model. After m iterations, the model with the largest number of interior points is selected as the final fitting result.

[0033] In m iterations, under the given confidence probability, the minimum data size of the fitted model and the proportion of in-data points P (P > t) satisfy the following relationship:

[0034] P = 1 - (1 - t) N ) m

[0035]

[0036] Where P is the confidence level; t represents the proportion of points within the total data; N is the minimum data size for the fitted model; and 1-t represents the total data size. N When selecting N points, there must be at least one outside point; when When the iteration ends, the iteration is terminated.

[0037] The beneficial effects of this invention are as follows:

[0038] 1. This method can accurately extract highway pavement point clouds when it is impossible to use roadside features for boundary recognition, and provides a feasible and relatively complex solution for extracting roads without roadside features.

[0039] 2. A filtering algorithm chain method is used to denoise complex vehicle-mounted laser point cloud data, which improves the accuracy and efficiency of subsequent elevation correction processing and highway boundary line extraction.

[0040] 3. The elevation accuracy of point cloud data is corrected by using cubic Hermit interpolation and inverse distance weighting constraint method. Compared with the previous method of setting up targets to improve the accuracy of point cloud data, this method can better ensure the accuracy of point cloud data and enable point cloud data to meet engineering applications. Attached Figure Description

[0041] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, use the same reference numerals to denote the same or similar parts. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0042] Figure 1 This is a flowchart of one embodiment of the present invention;

[0043] Figure 2 This is a point cloud data image after Gaussian filtering;

[0044] Figure 3 This is a point cloud data image after surface fitting and filtering.

[0045] Figure 4 This is a point cloud data image after normal vector filtering;

[0046] Figure 5 This is a point cloud data image after morphological filtering;

[0047] Figure 6 This is a point cloud data image after process correction;

[0048] Figure 7 The resulting road map is extracted from the highway boundary. Detailed Implementation

[0049] like Figure 1 The highway boundary extraction method based on vehicle-mounted laser point clouds shown includes:

[0050] S1: Collect point cloud data of roads and features on both sides of the roads, and preprocess the point cloud data;

[0051] S2: Obtain the control points corresponding to the point cloud data, classify the control points to obtain the shape point data; then perform three Hemite interpolation processes on the shape point data; finally, use IDW to allocate the elevation difference between the interpolated shape point data and the point cloud data, and perform elevation correction processing on the point cloud data.

[0052] S3: First, the DBSCAN clustering algorithm combined with density and reflection intensity filtering is used to extract road point cloud data from the point cloud data after elevation correction; then, the Alpha Shape edge detection algorithm is used to extract boundary points from the extracted highway point cloud data; finally, the RANSC line fitting method is used to fit the highway boundary points to obtain the complete highway boundary.

[0053] According to one embodiment of this application, the preprocessing of point cloud data in step S1 above includes:

[0054] 1. Construct a KD tree point cloud diagram based on the collected point cloud data.

[0055] KD-trees (K-Dimensional Trees) are a structure first introduced by Bentleyz for multidimensional spatial data. By partitioning multidimensional data, they significantly improve the efficiency of neighborhood search. Constructing a KD-tree is equivalent to repeatedly dividing k-dimensional space using hyperplanes perpendicular to the coordinate axes. This partitioning forms a tree structure where each node represents a hyperplane partition, and leaf nodes contain the actual data points in the space. When a partition point is selected in a dimension, a hyperplane perpendicular to the coordinate axes is drawn in that dimension, dividing the space into two parts. This partition point divides the space into left and right subspaces. The left subspace contains all points less than or equal to the partition point in that dimension, and the right subspace contains all points greater than the partition point. Then, when selecting the next dimension, the above process is repeated. By continuously selecting partition points in each dimension, a KD-tree can be constructed. This partitioning process can be viewed as dividing k-dimensional space into two parts at each node; therefore, the construction of a KD-tree is essentially a binary tree construction process. By constructing a KD tree-structured point cloud graph, the efficiency of neighborhood search can be improved and the computation time reduced during the processing of point cloud data.

[0056] 2. Then, the point cloud data is denoised.

[0057] This application employs a point cloud denoising method combining Gaussian filtering, surface fitting filtering, normal vector filtering, and morphological filtering to denoise vehicle-mounted laser point cloud data. Multiple filtering algorithms are combined in a specific order to process the point cloud data sequentially. Each filtering algorithm in the chain further optimizes and improves the result processed by the previous algorithm. In Gaussian filtering, a Gaussian function is used as a weighting function to smooth the point cloud data, helping to reduce noise while preserving the main features of the point cloud; the point cloud data after Gaussian filtering is shown in the image below. Figure 2As shown. The basic principle of surface fitting filtering is to fit a complex spatial surface using several simple quadratic surfaces; by setting an appropriate threshold, noise points can be effectively identified and removed, making the fitted surface smoother and more continuous; the point cloud data after surface fitting filtering is shown in the figure. Figure 3 As shown. Normal vector filtering calculates the normal information of point cloud data using normal estimation methods. A normal is the normal vector at each point on the point cloud surface, which describes the orientation and shape of the point cloud surface. Then, a threshold for normal curvature is set to describe the curvature of the point cloud surface. Curvature represents the degree of bending of the point cloud surface at that point; the normal curvature is the curvature at that point. The point cloud data after normal vector filtering is shown in the image. Figure 4 As shown. Mathematical morphological filtering uses morphological opening operations to process the profile data, and then further optimizes the results through adaptive regression. The point cloud data before and after mathematical morphological filtering are shown in the figure. Figure 5 As shown.

[0058] 3. Finally, the denoised point cloud data is simplified.

[0059] High-precision vehicle-mounted laser 3D scanning technology generates massive point cloud data during data acquisition. Each point carries location information (X, Y, Z) as well as additional information such as color and intensity. This data volume is enormous, requiring a significant amount of storage space. Simplifying point cloud data can reduce the required storage space, effectively reducing the computational burden and improving the efficiency of data processing, analysis, and rendering. In some applications, not all point cloud data is useful. For example, in topographic mapping, only ground points may be of interest, while points above vegetation or buildings may be irrelevant. Simplifying the data removes unnecessary information, allowing resources to be more focused on processing and analyzing key data. In some cases, excessive point cloud data may interfere with the accuracy of analysis due to noise or overlap. Simplifying and cleaning point cloud data reduces noise, highlights useful information, and thus improves the accuracy of data analysis. This application employs an octree point cloud simplification method to simplify point cloud data, specifically including: first, processing the denoised point cloud data and dividing it into cubes. The cubes constitute a hierarchical spatial structure, namely an octree. Within each small cube, the point closest to the cube's centroid is selected as the representative point, replacing all other points. This method simplifies the point cloud data, reducing the amount of data. The degree of simplification is determined by the size of each cube in the octree. Typically, the degree of simplification can be controlled by adjusting the size of the smallest cube. Smaller cube sizes result in finer simplification, while larger cube sizes produce coarser simplification. Using an octree-based point cloud simplification algorithm to simplify the experimental point cloud ensures the efficiency and accuracy of subsequent elevation correction and highway boundary extraction processing.

[0060] According to one embodiment of this application, step S2, performing three Hemite interpolation processes on the type value point data, includes:

[0061] 1. The parameterization of the model points is performed using the standardized cumulative chord length method to determine the nodal function values;

[0062] 2. Calculate the shear loss using Bézier curves to determine the nodal derivative values;

[0063] 3. Select a set of interpolation points, where each interpolation point includes a nodal function value and a nodal derivative value; these interpolation points are usually specific points in the original data and can be selected according to the requirements of the problem or the characteristics of the data;

[0064] 4. Construct interpolation basis functions, and then linearly combine the interpolation basis functions with the corresponding function values ​​and derivative values ​​to obtain a cubic polynomial;

[0065] Hermite basis functions can be used for interpolation. Hermite basis functions possess certain properties that satisfy the requirements for function values ​​and derivative values ​​needed for interpolation. Hermite basis functions can be expressed as:

[0066] H0(x)=(1-3t 2 +2t 3 )

[0067] H1(x)=t(1-t 2 )

[0068] H2(x)=t 2 (3-2t)

[0069] H3(x)=t 2 (1-t)

[0070] Where t is a parameter ranging from [0, 1], it represents the position of the interpolation between two adjacent interpolation points. In actual calculations, the difference between the independent variable x and the interpolation point position can be divided by the interpolation interval length, i.e., t = (x, y) / x. i ) / h, where is the position of the interpolation point and h is the interpolation interval length.

[0071] The general form of the interpolating polynomial is:

[0072] P(x) = H0(x)·y i +H1(x)·h·dy i / dx i +H2(x)·y {i+1} +H3(x)·h·dy {i+1} / dx {i+1}

[0073] Among them, y i and y {i+1} Let d represent the function values ​​at two adjacent interpolation points. yi / d xi and dy {i+1} / dx {i+1} These represent the derivative values ​​at the two interpolation points, respectively.

[0074] By calculating the linear combination of the interpolation basis functions, their corresponding function values, and derivative values, we obtain a cubic polynomial. This polynomial is called the cubic Hermitian interpolation polynomial, denoted as P(x).

[0075] P (x) =H0(x)·y0+H1(x)·m0+H2(x)·y1+H3(x)·m1

[0076] Here, y0 and y1 represent the function values ​​at two adjacent interpolation points, m0 and m1 represent the derivative values ​​at these two interpolation points, and H0(x), H1(x) and H2(x) are the Hermite basis functions mentioned earlier.

[0077] 4. Calculate the interpolation curve or surface between interpolation points using a cubic polynomial. By calculating the interpolation polynomial P(x), the interpolation curve or surface between interpolation points can be obtained. For one-dimensional interpolation problems, the interpolation polynomial can be used to calculate the predicted function value at a given independent variable x. For two-dimensional or higher-dimensional interpolation problems, the interpolation polynomial can be applied to the corresponding surface or hypersurface.

[0078] After three Hermitian interpolation operations, the final encrypted control point results are obtained. Nearest neighbor search is used to find the nearest point to each control point. Theoretically, the nearest neighbor should coincide with the control point. However, due to the inherent errors in the data acquired by the vehicle-mounted laser scanning system, elevation correction of the point cloud data is performed by allocating the elevation difference between the nearest neighbor and the control point using IDW (Integrated Dynamic Wavelength Wrapper). Assume the control point is A, and its nearest neighbor is A''. 近 Based on distance, the nearest neighbor point A is divided. 近 The points are A1, A2, A3...A n A can be obtained by calculating the Euclidean distance. 近 With A n The distance is:

[0079]

[0080] The elevation difference between the nearest neighbor point and the control point should be:

[0081] D 差 =Z 控 -Z 近

[0082] The formula for calculating the weight based on distance is:

[0083]

[0084] At different distances, the weight constraint of a control point on its nearest neighbor is smaller. Therefore, the farther away a point is, the smaller its impact on the elevation correction of a single control point. However, as mentioned in the previous section, interpolation is performed on the control points to ensure that there are enough control points to assign weights to points that are farther away, thus achieving the purpose of elevation correction. By summing the weights of all points, the total weight w can be obtained. 总 .

[0085] Ultimately, a single control point is used to improve the local accuracy of the laser point cloud data. The overall control point, through the improvement of local accuracy, affects the overall accuracy, thus achieving the goal of improving overall accuracy. The point cloud data after elevation correction is shown below. Figure 6 As shown.

[0086] According to one embodiment of this application, step S2 further includes longitudinal and lateral verification of the point cloud data before and after elevation correction processing. Specifically, before and after the elevation correction algorithm optimization of the point cloud data, the elevation values ​​of the ten closest points to the verification point are selected using the K-value nearest neighbor search method. Then, the obtained elevation values ​​are compared with the verification points (i.e., the true values) to determine whether the accuracy of the point cloud data has been effectively improved. By comparing and analyzing the accuracy optimization results of the point cloud data in the longitudinal and lateral directions, the accuracy of the subsequent highway boundary extraction results can be guaranteed.

[0087] (1) Vertical verification of the accuracy of point cloud data after elevation correction: Since the verification point and the model point are on the same vertical line, the elevation of the verification point is taken as the true value to verify the vertical accuracy of the point cloud data after elevation correction.

[0088] (2) Lateral verification of point cloud data accuracy after elevation correction: When extracting the highway boundary, the highway boundary is in the lateral direction of the model point. However, since the verification point and the model point are on the same vertical line, when performing lateral verification, one segment of model point data is retained from the four segments of model point data used for elevation correction for lateral accuracy verification.

[0089] According to one embodiment of this application, step S3, extracting road point cloud data from the elevation-corrected point cloud data using the DBSCAN clustering algorithm combined with density and reflection intensity filtering, includes:

[0090] 1. The DBSCAN clustering algorithm is used to distinguish points on the road surface from points on other non-road surfaces; the specific implementation steps of DBSCAN clustering are as follows:

[0091] (1) Initialization: Select appropriate radius Eps and minimum neighborhood point number MinPts parameters, and mark all points in the point cloud as unvisited.

[0092] (2) Random point selection and classification: Randomly select an unvisited point and check the number of points in its Eps neighborhood. If the number of points in the neighborhood is greater than or equal to the number of points in the neighborhood, mark it as a core point and search for all densely reachable points that can be connected into a cluster. Otherwise, if the condition for being a core point is not met, mark it as a boundary point or a noise point.

[0093] (3) Neighborhood expansion: For each point in the neighborhood of a core point, repeat the above process to check whether they are also core points, thereby recursively expanding the cluster.

[0094] (4) Cluster construction: Continue the above process until all points have been visited. Finally, each core point and the points that are density-reachable with it form a unique cluster, while noise points are excluded.

[0095] 2. Points on the road surface are distinguished from points on other non-road surfaces based on their point cloud density. When using a vehicle-mounted laser scanning system to collect data on highways, the point cloud density in the direction perpendicular to the highway gradually decreases as the distance from the laser scanner increases. This phenomenon results in a higher point cloud density in highway areas compared to non-highway areas, even though the relative distance between the scanning system and the ground remains roughly constant. This setup ensures a uniform distribution of ground point cloud density at the highway's edge. Using the KD-tree indexing method, for any given point P, the number of point clouds *n* within a region centered at P with radius *R* is calculated, thus estimating the point cloud density at point P. After traversing the entire point cloud dataset, point clouds with a density greater than a threshold are identified as highway points; otherwise, they are considered highway points.

[0096] 3. Distinguish points on road surfaces from points on other non-road surfaces based on differences in point cloud reflectivity. When identifying highway surfaces, highway surfaces have higher reflectivity compared to non-highway surfaces, which can be extracted using the reflectivity intensity value. However, in laser scanning, the brightness intensity of the point cloud decreases with increasing distance due to the difference in distance and reflection angle between the highway and the scanner. To reduce the impact of these factors on the brightness intensity of the point cloud, a common method is to standardize the brightness intensity of the highway surface according to distance. This method allows the calculation of the maximum value of the standardized brightness intensity, which is then used as a threshold. In road surface point cloud extraction, if the standardized intensity value of a point falls within the threshold range, that point is considered a highway surface point. This process simplifies the original description using more technical language, aiming to reduce its complexity. Similarly, the threshold can be determined through experiments on sample road sections.

[0097]

[0098] Where d represents the distance from the road surface to the scanning device, d mean I represents the average distance, and I represents the unnormalized reflection intensity value. n This represents the normalized reflection intensity value.

[0099] According to one embodiment of this application, step S3, extracting boundary points from the extracted highway point cloud data using the AlphaShape edge detection algorithm, includes:

[0100] 1. Perform Delaunay triangulation on the input point set to generate a triangular network that connects all input points and satisfies the Delaunay condition;

[0101] 2. Based on the constructed Delaunay triangulation, each triangle or edge is evaluated, its circumcircle radius is calculated, and it is compared with a preset α value. If the circumcircle radius of a triangle or edge is less than or equal to the α value, the triangle or edge is considered to be part of the target shape; if the radius is greater than the α value, the triangle or edge is considered not to belong to the target shape and must be discarded.

[0102] 3. Through the above filtering steps, the remaining triangles or edges constitute the target shape that meets the Alpha Shapes condition; based on this, the edges or boundaries of the shape are extracted to obtain the outline of the target object.

[0103] Alpha Shape offers advantages such as preserving road shape, robustness to noise points, adjustable parameters, and high computational efficiency. It can adaptively extract the convex hull region of the road boundary based on the point set distribution, while effectively ignoring noise points, thus improving the accuracy and efficiency of extraction.

[0104] According to one embodiment of this application, in step S3, the RANSC straight line fitting method is used to fit the highway boundary points to obtain the complete highway boundary, including:

[0105] A subset of point cloud data is selected through random sampling. A model is fitted to this subset, and this model is compared with all point cloud data. A pre-set threshold is used to determine which data points fit the model, thus obtaining a more reliable model. After m iterations, the model with the largest number of interior points is selected as the final fitting result. Figure 7 As shown;

[0106] In m iterations, under the given confidence probability, the minimum data size of the fitted model and the proportion of in-data points P (P > t) satisfy the following relationship:

[0107] P = 1 - (1 - t) N ) m

[0108]

[0109] Where P is the confidence level; t represents the proportion of points within the total data; N is the minimum data size for the fitted model; and 1-t represents the total data size. N When selecting N points, there must be at least one outside point; when When the iteration ends, the iteration is terminated.

[0110] By adjusting the distance threshold of the RANSC fitting algorithm, it was ensured that the fitting results for the highway boundary did not include too many outliers. Accurate setting of the algorithm's iteration count allowed it to find a suitable model. The final fitting results for the highway boundary points also verified the robustness of the RANSC algorithm.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for extracting highway boundaries based on vehicle-mounted laser point clouds, characterized in that, include: S1: Collect point cloud data of the road and the features on both sides of the road, and preprocess the point cloud data; S2: Obtain control points corresponding to the point cloud data, classify the control points to obtain shape point data; then perform three Hemite interpolation processes on the shape point data; finally, use IDW to allocate the elevation difference between the interpolated shape point data and the point cloud data, and perform elevation correction processing on the point cloud data. S3: First, the DBSCAN clustering algorithm combined with density and reflection intensity filtering is used to extract road point cloud data from the point cloud data after elevation correction; then, the Alpha Shape edge detection algorithm is used to extract boundary points from the extracted highway point cloud data; finally, the RANSC line fitting method is used to fit the highway boundary points to obtain the complete highway boundary. The method of extracting road point cloud data from elevation-corrected point cloud data using the DBSCAN clustering algorithm combined with density and reflection intensity filtering includes: The DBSCAN clustering algorithm is used to distinguish points on the road surface from points on other non-road surfaces; Points on the road surface are distinguished from points on other non-road surfaces based on the different point cloud densities; Points on the road surface are distinguished from points on other non-road surfaces based on the different reflectivities of the point cloud; The step of extracting boundary points from the extracted highway point cloud data using the AlphaShape edge detection algorithm includes: Perform Delaunay triangulation on the input point set to generate a triangulation that connects all input points and satisfies the Delaunay condition; Based on the constructed Delaunay triangulation, each triangle or edge is evaluated, its circumcircle radius is calculated, and it is compared with a preset α value. If the circumcircle radius of a triangle or edge is less than or equal to the α value, the triangle or edge is considered to be part of the target shape; if the radius is greater than the α value, the triangle or edge is considered not to belong to the target shape and must be discarded. Through the above filtering steps, the remaining triangles or edges constitute the target shape that meets the Alpha Shapes condition; based on this, the edges or boundaries of the shape are extracted to obtain the outline of the target object.

2. The highway boundary extraction method based on vehicle-mounted laser point clouds according to claim 1, characterized in that, In step S1, the preprocessing of the point cloud data includes: Construct a KD tree-shaped point cloud map based on the collected point cloud data; Denoising the point cloud data; The denoised point cloud data is then simplified.

3. The highway boundary extraction method based on vehicle-mounted laser point clouds according to claim 2, characterized in that, The denoising process for point cloud data includes sequentially applying Gaussian filtering, surface fitting filtering, normal vector filtering, and morphological filtering to denoise the point cloud data.

4. The highway boundary extraction method based on vehicle-mounted laser point clouds according to claim 1, characterized in that, Step S2, performing three Hemite interpolation processes on the type value point data, includes: The parameterization of the model points is performed using the standardized cumulative chord length method to determine the nodal function values; Shear loss is calculated using Bézier curves to determine nodal derivative values; Select a set of interpolation points, wherein each interpolation point in the set includes the nodal function value and the nodal derivative value; By constructing interpolation basis functions and then linearly combining the interpolation basis functions with the corresponding function values ​​and derivative values, a cubic polynomial can be obtained. The curve or surface for interpolation between interpolation points is calculated based on the cubic polynomial.

5. The highway boundary extraction method based on vehicle-mounted laser point clouds according to claim 1, characterized in that, Step S2 further includes performing longitudinal and lateral verification on the point cloud data before and after elevation correction processing.

6. The highway boundary extraction method based on vehicle-mounted laser point clouds according to claim 1, characterized in that, In step S3, the RANSC straight line fitting method is used to fit the highway boundary points to obtain the complete highway boundary, which includes: A subset of point cloud data is selected by random sampling, a model is fitted on this subset, and the model is compared with all point cloud data. Based on a pre-set threshold, it is determined which data points fit the model, thus obtaining a more reliable model. After m iterations, the model with the largest number of interior points is selected as the final fitting result. In m iterations, under the given confidence probability, the minimum data size of the fitted model and the proportion of in-data points P (P > t) satisfy the following relationship: P=1-(1-t N ) m Where P is the confidence level; t represents the proportion of points within the total data; N is the minimum data size for the fitted model; and 1-t represents the total data size. N When selecting N points, there must be at least one outside point; when When the iteration ends, the iteration is terminated.

Images