Pavement point cloud extraction method and system based on facade point detection

Abstract

The invention discloses a road surface point cloud extraction method and system based on facade point detection, and belongs to the technical field of point cloud processing, and the method comprises the steps: obtaining a road point cloud, and carrying out the preprocessing of the road point cloud based on a driving track; facade points in the road point cloud are detected and removed; screening the road point clouds with the facade points removed based on normal vector features to form preliminary road surface point clouds; and performing pavement point cloud and interference point separation on the preliminary pavement point cloud through density clustering, further optimizing the clustered pavement point cloud by applying a cloth simulation filtering algorithm, and outputting a clean pavement point cloud. According to the method, the road edge point cloud can be effectively reserved, the integrity and precision of the road point cloud are improved, and the requirements of applications such as road high-precision modeling and automatic driving are met.

Description

Technical Field

[0001] The present invention belongs to the technical field of point cloud processing, and in particular relates to a road surface point cloud extraction method and system based on facade point detection. Background Art

[0002] In recent years, mobile laser scanning (MLS) technology has rapidly developed and has become an important means of obtaining high-precision three-dimensional information about roads and their surroundings. By integrating devices such as LiDAR, Global Navigation Satellite System (GNSS), and Inertial Measurement Unit (IMU) onto a mobile platform, the system continuously collects high-density, high-precision, and real-time 3D point cloud data of roads and their surroundings while the vehicle is in motion. This data is widely used in a variety of fields, including urban 3D modeling, road facility management, and autonomous driving map construction, providing strong support for the digital management of urban infrastructure and the development of intelligent transportation systems.

[0003] In the point cloud data processing process, road surface point cloud extraction, as the foundation for subsequent modeling and information extraction, has a significant impact on the overall processing results. High-quality road surface point clouds must possess good continuity and integrity to support subsequent road modeling and high-precision map production.

[0004] In existing research, road surface point cloud extraction methods can be mainly classified into the following categories:

[0005] ① Scanline method: This method divides the point cloud into scan lines based on the order of laser scanning and extracts road surface points by analyzing features such as elevation and slope along the scan lines. This method performs well in regular road scenes, but extraction accuracy is easily affected in complex urban environments due to the irregular scan line structure and complex road types.

[0006] ② Cluster analysis: Based on the feature similarity of point clouds within local areas, clustering algorithms such as region growing and DBSCAN are used to classify point clouds into different categories, thereby identifying road surface areas. However, in scenes with a large number of non-road surface points (such as vehicles, pedestrians, and vegetation), clustering results are easily disturbed, affecting the accuracy of road surface point cloud extraction.

[0007] ③ Machine learning: This method constructs a point cloud feature vector containing elevation, normal vectors, and multi-scale features, and uses classifiers such as support vector machines to classify the point cloud, thereby extracting road surface points. This method has high accuracy when there is sufficient training data, but the model's generalization ability is limited in new scenarios or heterogeneous data sources, and it is highly dependent on the training samples.

[0008] ④ Point cloud rasterization: Projecting a 3D point cloud onto a 2D plane generates a raster image. Road feature information is then extracted using image processing techniques such as threshold segmentation and edge detection. This method is computationally efficient, but can lose spatial structural information during the projection process, particularly around road edges. This can lead to jagged errors caused by grid resolution limitations, compromising extraction accuracy and integrity.

[0009] In summary, existing road point cloud extraction methods have problems such as insufficient accuracy, poor robustness, and strong dependence on training data when dealing with complex urban road environments. In particular, they are prone to missing points in road edge areas. Summary of the Invention

[0010] In response to the deficiencies in the prior art, the present invention provides a road surface point cloud extraction method and system based on facade point detection, which can effectively retain road edge point clouds and improve the integrity and accuracy of road point clouds to meet the needs of applications such as high-precision road modeling and autonomous driving.

[0011] The present invention provides the following technical solutions:

[0012] In a first aspect, a road surface point cloud extraction method based on facade point detection is provided, comprising:

[0013] Obtain road point cloud and pre-process it based on driving trajectory;

[0014] Detect and remove vertical points in the road point cloud;

[0015] The road point cloud with facade points removed is filtered based on normal vector features to form a preliminary road surface point cloud;

[0016] The preliminary road surface point cloud is separated from the interference points through density clustering, and the clustered road surface point cloud is further optimized by applying the cloth simulation filtering algorithm to output a clean road surface point cloud.

[0017] Optionally, the pre-processing of the road point cloud based on the driving trajectory includes:

[0018] Based on the timestamp, the road point cloud is matched with the driving trajectory points, and the point cloud outside the height range and the point cloud far away from the vehicle's driving trajectory are eliminated by straight-through filtering.

[0019] Optionally, when excluding point clouds outside the height range, the height range is set according to the trajectory point elevation H, the vehicle-mounted acquisition device installation height h0, and the tolerance d0. The specific height range is: [H-h0-d0,H];

[0020] When removing point clouds that are far away from the vehicle's driving trajectory, a horizontal distance threshold is set between each point in the point cloud data and the matched trajectory point. Points exceeding the threshold are considered to be far away from the vehicle's driving trajectory.

[0021] Optionally, the detecting and removing of elevation points in the road point cloud is specifically as follows:

[0022] For each point in the road point cloud, a cylindrical neighborhood with a radius of r and a height of h is established, and the elevation values ​​of each point in the neighborhood are extracted and arranged in ascending order;

[0023] All points in the neighborhood whose elevation values ​​are between the 25% and 75% quantiles are selected as sample points, and the sample points are used for local plane fitting;

[0024] If the angle between the normal vector of the fitted plane and the horizontal plane is less than the preset threshold, the points in the neighborhood whose distance to the fitted plane is less than the threshold are marked as vertical points and removed.

[0025] Optionally, the method of using sample points to perform local plane fitting is to use the RANSAC method to fit the optimal plane Ax+By+Cz+D=0, then the normal of the fitting plane is (A, B, C), and the unit normal vector of the fitting plane is

[0026] Optionally, when establishing the cylindrical neighborhood of point p, if the number of points in the cylindrical neighborhood is less than a set threshold, or when the number of sample points in the neighborhood of point p is less than a set threshold, point p is skipped and the cylindrical neighborhoods of other points in the road point cloud are re-established.

[0027] Optionally, the road point cloud with facade points removed is screened based on the normal vector feature to form a preliminary road surface point cloud, specifically:

[0028] For the point cloud after removing the facade points, the normal vector of each point is calculated based on the neighboring points in the spherical neighborhood, and the points whose angle between the normal vector and the vertical direction is less than the set threshold are screened out to form a preliminary road surface point cloud;

[0029] The separation of the preliminary road surface point cloud and interference points by density clustering is specifically as follows:

[0030] The DBSCAN clustering method based on density connectivity is used to divide the preliminary point cloud into multiple clusters according to the set neighborhood distance and minimum cluster capacity parameters. The cluster with the largest number of points is selected as the road surface point cloud, and other clusters are eliminated.

[0031] In a second aspect, a road surface point cloud extraction system based on facade point detection is provided, comprising:

[0032] The preprocessing module obtains the road point cloud and preprocesses the road point cloud based on the driving trajectory;

[0033] The facade point processing module detects and removes facade points in the road point cloud;

[0034] The preliminary extraction module filters the road point cloud excluding the facade points based on the normal vector features to form a preliminary road surface point cloud;

[0035] The extraction module is optimized to separate the preliminary road surface point cloud from interference points through density clustering, and the clustered road surface point cloud is further optimized by applying the cloth simulation filtering algorithm to output a clean road surface point cloud.

[0036] According to a third aspect, a computer device is provided, comprising a processor and a memory; wherein, when the processor executes a computer program stored in the memory, the steps of the road surface point cloud extraction method based on facade point detection as described in any one of the first aspects are implemented.

[0037] In a fourth aspect, a computer-readable storage medium is provided for storing a computer program; when the computer program is executed by a processor, the steps of the road surface point cloud extraction method based on facade point detection described in any one of the first aspects are implemented.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The detection and elimination of vertical points in the present invention can effectively eliminate the interference caused by vertical surfaces such as road bumps and guardrails, avoid misjudging non-road surface points as road surface, and completely preserve the continuous point cloud of the road edge, thereby improving the accuracy and integrity of road surface extraction. In addition, the present invention effectively suppresses environmental noise and isolated outliers through multi-stage filtering - including vertical point elimination, normal vector screening, density clustering and cloth simulation filtering, and realizes high-quality, clean and continuous road surface point cloud extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of the road surface point cloud extraction method based on facade point detection of the present invention.

[0041] Figure 2 It is the test road section point cloud data of the present invention.

[0042] Figure 3 This is a point cloud data preprocessing result diagram of the present invention.

[0043] Figure 4 This is the point cloud result image after facade point detection and elimination of the present invention.

[0044] Figure 5 It is a schematic diagram of the interference of the normal vector at the junction of the road surface and the facade of the present invention.

[0045] Figure 6Schematic diagram of the normal vector improvement after removing the facade points.

[0046] Figure 7 Schematic diagram of candidate road point clouds screened based on normal vectors.

[0047] Figure 8 This is the point cloud map of the main road area extracted based on density clustering.

[0048] Figure 9 The final road surface point cloud result after cloth simulation filtering optimization.

[0049] Figure 10 The cross-sectional comparison diagram of the original point cloud and the extracted road surface point cloud. DETAILED DESCRIPTION

[0050] The present invention will be further described below with reference to the accompanying drawings. The following examples are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. It should be noted that the term "comprising" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0051] Example 1

[0052] like Figure 1 As shown, a road surface point cloud extraction method based on facade point detection is provided, comprising the following steps:

[0053] S1: Obtain road point cloud and preprocess the road point cloud based on the driving trajectory.

[0054] The road point cloud includes the GPSTime, X, Y, and Z field values ​​of each point, and the driving trajectory includes the GPSTime, X, Y, and Z field values ​​of each trajectory point.

[0055] Preprocessing of road point clouds is to use vehicle driving trajectories to preliminarily screen the point clouds and retain point cloud data that may belong to the road range.

[0056] Preprocessing specifically involves matching each point in the vehicle-mounted laser point cloud with the vehicle trajectory point at the corresponding time using the GPSTime field value. The matching method can refer to existing technologies. The vehicle-mounted device installation height is set to h0, the tolerance is d0, and the trajectory point elevation is H. A reasonable height range is defined as [H-h0-d0, H], and point clouds outside this height range are removed using a straight-through filter.

[0057] Set the horizontal distance threshold d between each point in the point cloud data and the matching trajectory point t , calculate each point in the point cloud (x i ,y i ,z i ) to the corresponding trajectory point (x t ,y t ,z t ) horizontal distance If d xy >d t , then it is determined that the point is too far away from the vehicle, that is, the point is far away from the vehicle's driving trajectory and is removed.

[0058] S2: Detect and remove vertical points in the road point cloud.

[0059] For vertical structures such as curbs and guardrails at the edge of the road, their corresponding facade points are detected and removed to avoid insufficient road surface extraction accuracy caused by the connection between facade points and road surface points.

[0060] Step S2 specifically includes:

[0061] S21: With each point in the point cloud as the center, a cylindrical neighborhood with a radius of r and a height of h is established, and the points in the neighborhood are searched. If the number of points in the neighborhood is less than the set threshold, the current point is skipped.

[0062] S22: Extract the elevation values ​​of all points in the neighborhood and arrange them in ascending order to determine the 25% quantile elevation H1 and the 75% quantile elevation H2.

[0063] S23: Select points whose elevations are in the interval [H1, H2] as the sample point set. If the number of sample points is too small, skip it.

[0064] S24: Use the RANSAC method to fit the optimal plane Ax+By+Cz+D=0 for the sample point set, then the normal of the plane is (A, B, C), and calculate the unit normal vector of the plane

[0065] S25: Calculate the angle θ between the normal vector of the fitted plane and the horizontal plane = arccos(|n e (0,0,1)|), if the angle θ is less than the preset threshold θ0, the plane is considered to be approximately perpendicular to the ground and can be judged as a facade structure, and the facade point detection continues. Otherwise, the current point is skipped.

[0066] S26: Take each point (x, y, z) in the cylindrical neighborhood, and place the elevation z in the interval (H1, H2) and the distance to the fitting plane Points smaller than the threshold are marked as facade points.

[0067] S27: Eliminate all point clouds marked as facade points to obtain point cloud data without facade interference.

[0068] S3: The road point cloud after removing the facade points is filtered based on the normal vector features to form a preliminary road surface point cloud.

[0069] Taking advantage of the fact that the road surface is approximately flat, non-road points are further eliminated based on the normal vector.

[0070] S31: For the point cloud after removing the facade points, the normal vector of each point is calculated based on the neighboring points in the spherical neighborhood.

[0071] Specifically, a local plane fitting is performed on the points in the neighborhood, and the normal vector of the fitted plane is used as the normal vector of the point. The fitting methods that can be used include least squares method, principal component analysis, etc.

[0072] S32: Calculate the angle θ between the normal vector of each point and the vertical direction = arccos(|n e (0,0,1)|), retaining points with angles smaller than the set threshold to form a preliminary road surface point cloud.

[0073] S4: Separate the preliminary road surface point cloud from the interference points through density clustering, and further optimize the clustered road surface point cloud using the cloth simulation filtering algorithm to output a clean road surface point cloud.

[0074] S41: The DBSCAN clustering method based on density connectivity is used to divide the preliminary point cloud into multiple clusters according to the set neighborhood distance and minimum cluster capacity parameters. The cluster with the largest number of points is selected as the road surface point cloud, and other clusters are eliminated.

[0075] S42: Set the parameters of the cloth simulation filter, including grid resolution, classification threshold, smoothing, time step, rigidity and number of iterations, and apply it to the clustered road point cloud;

[0076] S43: Based on the results of the filtering classification, only the point clouds determined to be ground points are retained, and a high-quality, continuous and clean road surface point cloud is output.

[0077] Density clustering is used to extract road surface areas to separate the road surface from other interference points. To further clean up residual interference points and improve the quality of the ground point cloud, a cloth simulation filtering algorithm is applied to the clustered point cloud to further optimize the road surface point cloud until a clean road surface point cloud is output. The density clustering method and cloth simulation filtering algorithm can refer to existing technologies.

[0078] This application addresses the problem in traditional methods where facade points interfere with normal vector calculation, resulting in incomplete extraction of road edge points. It proposes improving normal estimation accuracy by identifying and removing facade points. Furthermore, a complete extraction process combining facade point extraction, normal feature screening, density clustering, and fabric simulation filtering is constructed to achieve high-precision and high-completeness extraction of road surface point clouds, meeting the needs of applications such as urban road modeling and high-precision map construction.

[0079] Example 2

[0080] Provided is an example of extraction using the road surface point cloud extraction method of the present application.

[0081] The first step is to read the point cloud and trajectory data.

[0082] In this embodiment, the urban road segment point cloud data collected by the Z+F PROFILER 9012 vehicle-mounted laser scanner is selected as the test data. Figure 2 The selected road section is approximately 80 meters long, with minimal undulations and a single lateral width change. Pedestrians and vehicles obstruct the road surface, and embankments on both sides connect to the sidewalks. The road also includes typical urban features such as streetlights, road signs, roadside trees, and transformers, providing good representation. During acquisition, the vehicle traveled at approximately 25 km / h, and the spacing between adjacent scan lines was approximately 3.5 cm. The point cloud file was read, and the GPSTime, X, Y, and Z fields for each point were extracted.

[0083] The trajectory of the device is obtained by jointly calculating the GNSS receiver and the IMU. The trajectory file is read to obtain the GPSTime, X, Y, and Z field information of each trajectory point.

[0084] The second step is point cloud data preprocessing.

[0085] Through the GPSTime field value, each point in the vehicle-mounted laser point cloud is matched with the vehicle trajectory point at the corresponding time.

[0086] The height threshold is set based on the device installation height of 1.95m and the tolerance of 0.5m. The points within the range of 2.45m are retained downward from the scanner, that is, all points with Z values ​​lower than the Z coordinate z of the corresponding track point are retained. t And higher than (z t -2.45)m point cloud.

[0087] Eliminate distant points whose horizontal distance from the corresponding trajectory points exceeds 50m, and remove irrelevant point clouds that may come from non-road areas. The processing results are as follows: Figure 3 shown.

[0088] The third step is facade point detection and elimination.

[0089] With each point in the point cloud as the center, a cylindrical neighborhood with a radius of r = 0.2m and a height of h = 0.2m is constructed, and the point set within the neighborhood is searched. The Z values ​​of all points in the neighborhood are extracted and arranged in ascending order, and the 25% quantile elevation H1 and the 75% quantile elevation H2 are calculated. A sample point set with elevations between H1 and H2 is selected, and a local plane fitting is performed on it using the RANSAC algorithm. If the angle between the normal vector of the fitting plane and the horizontal plane is less than 25°, the area is judged to belong to a facade structure. All points with elevations between H1 and H2 and a distance to the fitting plane less than 1cm are marked as facade points and removed from the point cloud. The processing results are as follows: Figure 4 shown.

[0090] The fourth step is to filter candidate road surface points based on normal vectors.

[0091] In the point cloud after removing the facade points, a spherical neighborhood with a radius of 15cm is constructed with each point as the center, and its normal vector is calculated. In order to avoid the normal vector at the intersection being interfered by the facade, the facade points must be removed first. The effect is compared. Figure 5 、 Figure 6 shown.

[0092] Extract points where the angle between the normal vector and the vertical direction is less than 25° as candidate road points. The results are as follows: Figure 7 shown.

[0093] The fifth step is to extract the main road area by density clustering.

[0094] The density clustering algorithm DBSCAN is applied to the candidate road point cloud, with a cluster radius of 10 cm and a minimum cluster point number of 20 to extract a point set with spatial connectivity. The cluster with the largest number of points is retained as the main road area, and the remaining small clusters and discrete points are removed to remove scattered interference. The results are as follows: Figure 8 shown.

[0095] In the sixth step, cloth simulation filtering is applied to further optimize the results.

[0096] Based on the main road surface point cloud, the Cloth Simulation Filtering (CSF) algorithm is further introduced to remove noise and interference points above the road surface. The parameters are set as follows: grid resolution 0.1m, classification threshold 0.01m, rigidity 2, and maximum number of iterations 500. The final high-precision and well-continuous road surface point cloud result is output, such as Figure 9 shown.

[0097] Step 7: Road surface point cloud accuracy assessment.

[0098] This example evaluates the effectiveness of road point cloud extraction from three perspectives: accuracy, completeness, and precision. Accuracy represents the proportion of correct road points extracted, completeness represents the proportion of actual road points that were correctly extracted, and precision comprehensively reflects the overall situation of false extractions and missed extractions. After manual labeling and statistics, approximately 3.935 million road points were ultimately extracted. The specific evaluation results are shown in Table 1:

[0099] Table 1: Actual road surface point extraction

[0100]

[0101]

[0102] As shown in Table 1, the road point cloud extracted by this method has high accuracy and completeness, which can meet the requirements of high-precision road extraction. Figure 10 As shown in the figure, by comparing the original point cloud with the extracted road surface point cloud in a local cross-section near a road bump, it is clear that road surface points in edge areas are fully preserved. Furthermore, mis-extracted points are mainly concentrated at the edges of features, such as where cars, pedestrians, and the road meet. Missed points mainly occur in areas with rough or undulating roads, and some points fail to pass the screening due to large changes in their normal vectors.

[0103] Example 3

[0104] A road surface point cloud extraction system based on facade point detection, comprising:

[0105] The preprocessing module obtains the road point cloud and preprocesses the road point cloud based on the driving trajectory;

[0106] The facade point processing module detects and removes facade points in the road point cloud;

[0107] The preliminary extraction module filters the road point cloud excluding the facade points based on the normal vector features to form a preliminary road surface point cloud;

[0108] The extraction module is optimized to separate the preliminary road surface point cloud from interference points through density clustering, and the cloth simulation filtering algorithm is applied to further optimize the clustered road surface point cloud to output a clean road surface point cloud.

[0109] For more specific details about the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be described again here.

[0110] Example 4

[0111] The present invention provides a computer device comprising a processor and a memory; wherein, when the processor executes a computer program stored in the memory, the steps of the above-mentioned road surface point cloud extraction method based on facade point detection are implemented.

[0112] For more specific details of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0113] Example 5

[0114] The present invention provides a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the steps of the above-mentioned road surface point cloud extraction method based on facade point detection are implemented.

[0115] For more specific details of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0116] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. References to the same or similar parts between the various embodiments will be sufficient. The systems, devices, and storage media disclosed in the embodiments are described briefly because they correspond to the methods disclosed in the embodiments. For relevant details, refer to the method description.

[0117] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention or certain portions of the embodiments.

[0118] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A road surface point cloud extraction method based on facade point detection, characterized in that: include: Obtain road point cloud and pre-process it based on driving trajectory; Detect and remove vertical points in the road point cloud; The road point cloud with facade points removed is filtered based on normal vector features to form a preliminary road surface point cloud; The preliminary road surface point cloud is separated from the interference points through density clustering, and the clustered road surface point cloud is further optimized by applying the cloth simulation filtering algorithm to output a clean road surface point cloud.

2. The road surface point cloud extraction method based on facade point detection according to claim 1 is characterized in that: The pre-processing of the road point cloud based on the driving trajectory includes: Based on the timestamp, the road point cloud is matched with the driving trajectory points, and the point cloud outside the height range and the point cloud far away from the vehicle's driving trajectory are eliminated by straight-through filtering.

3. The road surface point cloud extraction method based on facade point detection according to claim 2 is characterized in that: When removing point clouds outside the height range, the height range is set according to the trajectory point elevation H, the vehicle acquisition equipment installation height h0 and the tolerance d0. The specific height range is: [H-h0-d0,H]; When removing point clouds that are far away from the vehicle's driving trajectory, a horizontal distance threshold is set between each point in the point cloud data and the matched trajectory point. Points exceeding the threshold are considered to be far away from the vehicle's driving trajectory.

4. The road surface point cloud extraction method based on facade point detection according to claim 1, characterized in that: The detection and removal of vertical points in the road point cloud is specifically as follows: For each point in the road point cloud, a cylindrical neighborhood with a radius of r and a height of h is established, and the elevation values ​​of each point in the neighborhood are extracted and arranged in ascending order; All points in the neighborhood whose elevation values ​​are between the 25% and 75% quantiles are selected as sample points, and the sample points are used for local plane fitting; If the angle between the normal vector of the fitted plane and the horizontal plane is less than the preset threshold, the points in the neighborhood whose distance to the fitted plane is less than the threshold are marked as vertical points and removed.

5. The road surface point cloud extraction method based on facade point detection according to claim 4 is characterized in that: The method of using sample points to perform local plane fitting is to use the RANSAC method to fit the optimal plane Ax+By+Cz+D=0, then the normal of the fitting plane is (A, B, C), and the unit normal vector of the fitting plane is 6. The road surface point cloud extraction method based on facade point detection according to claim 4 is characterized in that: When establishing the cylindrical neighborhood of point p, if the number of points in the cylindrical neighborhood is less than the set threshold, or when the number of sample points in the neighborhood of point p is less than the set threshold, point p is skipped and the cylindrical neighborhoods of other points in the road point cloud are re-established.

7. The road surface point cloud extraction method based on facade point detection according to claim 1 is characterized in that: The road point cloud is filtered based on the normal vector feature to remove the vertical points to form a preliminary road surface point cloud, specifically: For the point cloud after removing the facade points, the normal vector of each point is calculated based on the neighboring points in the spherical neighborhood, and the points whose angle between the normal vector and the vertical direction is less than the set threshold are screened out to form a preliminary road surface point cloud; The separation of the preliminary road surface point cloud and interference points by density clustering is specifically as follows: The DBSCAN clustering method based on density connectivity is used to divide the preliminary point cloud into multiple clusters according to the set neighborhood distance and minimum cluster capacity parameters. The cluster with the largest number of points is selected as the road surface point cloud, and other clusters are eliminated.

8. A road point cloud extraction system based on facade point detection, characterized in that: include: The preprocessing module obtains the road point cloud and preprocesses the road point cloud based on the driving trajectory; The facade point processing module detects and removes facade points in the road point cloud; The preliminary extraction module filters the road point cloud excluding the facade points based on the normal vector features to form a preliminary road surface point cloud; The extraction module is optimized to separate the preliminary road surface point cloud from interference points through density clustering, and the cloth simulation filtering algorithm is applied to further optimize the clustered road surface point cloud to output a clean road surface point cloud.

9. A computer device, characterized in that: The method comprises a processor and a memory; wherein, when the processor executes the computer program stored in the memory, the steps of the road surface point cloud extraction method based on facade point detection according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that Used to store computer programs; when the computer programs are executed by the processor, the steps of the road surface point cloud extraction method based on facade point detection according to any one of claims 1 to 7 are implemented.

Images