A method and apparatus for edge extraction based on point clouds

Abstract

This application provides a point cloud-based edge extraction method and apparatus. The method acquires a target point cloud with different coordinate feature points and detects the neighborhood information of the feature points using a voxel nearest neighbor search algorithm. Then, it acquires the difference features of the feature points based on the neighborhood information. These difference features include the smoothed displacement of the feature points and / or the maximum angle between adjacent projection vectors of the feature points. A discrimination threshold is calculated based on the difference features, and feature points with difference features greater than the discrimination threshold are marked as edge points. This method can acquire the neighborhood information of each feature point based on the voxel nearest neighbor search algorithm, improving the efficiency of neighborhood information acquisition; and it provides two methods for identifying edge points in the point cloud: smoothing processing or adjacent projection angles. Identifying edge points using the original point cloud data improves the accuracy of edge extraction, thereby enhancing the edge extraction effect.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method and apparatus for edge extraction based on point clouds. Background Technology

[0002] 3D point cloud edge extraction technology is a fundamental point cloud processing algorithm that can be applied to scenarios such as object contour extraction, object localization, and object measurement. Point cloud edges refer to edge measurement points that can represent the characteristics of an object, used to characterize its geometric features.

[0003] To accurately extract edge points from a large number of 3D points, depth maps can be used to identify edge points in the point cloud. This involves generating a depth map of the point cloud based on the distance of each point to the camera plane, calculating the height difference between each point based on the depth map, and then extracting points with a height difference greater than a certain threshold as edge points. However, the process of generating the depth map of the point cloud incurs some losses, affecting the accuracy of the depth map data and consequently reducing the precision of edge extraction. Summary of the Invention

[0004] This application provides a point cloud-based edge extraction method and apparatus to solve the problem of low accuracy in edge extraction.

[0005] In a first aspect, some embodiments of this application provide a point cloud-based edge extraction method, including:

[0006] Obtain a target point cloud, wherein the target point cloud includes feature points with different coordinates;

[0007] The neighborhood information of the feature points is detected using a voxel nearest neighbor search algorithm;

[0008] The difference features of the feature points are obtained based on the neighborhood information. The difference features include the smoothed displacement of the feature points and / or the maximum angle between adjacent projection vectors of the feature points.

[0009] Calculate the discrimination threshold based on the aforementioned difference characteristics;

[0010] Feature points whose difference features are greater than the discrimination threshold are marked as edge points.

[0011] In some embodiments of this application, the neighborhood information of the feature point is detected by a voxel nearest neighbor search algorithm, including: establishing a voxel grid based on the target point cloud and recording the index information of the voxel grid; calculating the grid index of the feature point according to the index information; querying the neighborhood grid of the feature point through the grid index; and obtaining the feature point information of the neighborhood grid.

[0012] In some embodiments of this application, obtaining the difference features of the feature points based on the neighborhood information includes: performing smoothing processing on the target point cloud according to the neighborhood information to generate a smooth point cloud; detecting the first coordinates of the target feature points in the smooth point cloud and detecting the second coordinates of the target feature points in the target point cloud; and calculating the coordinate difference between the first coordinates and the second coordinates.

[0013] In some embodiments of this application, the target point cloud is smoothed based on a bilateral filtering algorithm.

[0014] In some embodiments of this application, obtaining the difference features of the feature points based on the neighborhood information includes:

[0015] The target point cloud is fitted with a plane according to the neighborhood information, and outliers in the neighborhood feature points are removed by a preset distance threshold to generate a local fitting plane; the feature points are projected onto the local fitting plane to generate the projection vectors of the feature points; the angle between two adjacent projection vectors is calculated; and the maximum value of the angle is recorded.

[0016] In some embodiments of this application, outliers in a neighborhood feature point are removed by setting a preset distance threshold, including: querying the neighborhood feature points of the feature point based on the neighborhood information; detecting the target distance between the neighborhood feature points and the feature point; if the target distance is greater than the distance threshold, marking the neighborhood feature points as outliers and removing the outliers; if the target distance is less than or equal to the distance threshold, marking the neighborhood information points as inliers and fitting the local fitting plane based on the inliers.

[0017] In some embodiments of this application, calculating a discrimination threshold based on the difference features includes: obtaining the average value and standard deviation of the difference features; and calculating a discrimination threshold based on the average value and the standard deviation.

[0018] In some embodiments of this application, calculating a discrimination threshold based on the average value and the standard deviation includes: obtaining a threshold factor of the discrimination threshold; calculating the product of the threshold factor and the standard deviation; and calculating the sum of the product and the average value as the discrimination threshold.

[0019] In some embodiments of this application, marking feature points with feature values ​​greater than the discrimination threshold as edge points includes: traversing the feature points of the target point cloud to obtain the feature values ​​of the feature points; if the feature value is greater than the discrimination threshold, marking the feature point as the edge point; if the feature value is less than or equal to the discrimination threshold, marking the feature point as a non-edge point.

[0020] Secondly, some embodiments of this application also provide a point cloud-based edge extraction device, including a search module, a data processing module, and an edge point extraction module, wherein:

[0021] The search module is configured to acquire a target point cloud, which includes feature points with different coordinates; and to detect the neighborhood information of the feature points using a voxel nearest neighbor search algorithm.

[0022] The data processing module is configured to obtain the difference features of the feature points based on the neighborhood information, the difference features including the smoothed displacement of the feature points, and / or the maximum angle between adjacent projection vectors of the feature points; and calculate a discrimination threshold based on the difference features.

[0023] The edge point extraction module is configured to mark feature points whose difference features are greater than the discrimination threshold as edge points.

[0024] As can be seen from the above technical solutions, the edge extraction method and apparatus based on point clouds provided in some embodiments of this application can acquire target point clouds of feature points with different coordinates and detect the neighborhood information of feature points through a voxel nearest neighbor search algorithm. Then, the difference features of feature points are acquired based on the neighborhood information. These difference features include the smoothed displacement of the feature points and / or the maximum angle between adjacent projection vectors of the feature points. A discrimination threshold is calculated based on the difference features, and feature points with difference features greater than the discrimination threshold are marked as edge points. This method can acquire the neighborhood information of each feature point based on the voxel nearest neighbor search algorithm, improving the efficiency of acquiring neighborhood information; and it provides two methods for identifying edge points in the point cloud: smoothing processing or adjacent projection angles. Identifying edge points using the original point cloud data can improve the accuracy of edge extraction, thereby improving the effect of edge extraction. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a point cloud-based edge extraction method provided for some embodiments of this application;

[0027] Figure 2 Example diagrams of voxel grids and neighborhood grids for voxel nearest neighbor search provided in some embodiments of this application;

[0028] Figure 3This is a schematic diagram illustrating the process of obtaining the smoothed displacement provided in some embodiments of this application;

[0029] Figure 4a This is a schematic diagram illustrating the effect of internal folding edge points provided in some embodiments of this application;

[0030] Figure 4b This application provides schematic diagrams illustrating the effects of high curvature edge points in some embodiments.

[0031] Figure 4c This is a schematic diagram illustrating the effect of outer contour boundary points provided in some embodiments of this application;

[0032] Figure 5 Example diagram of the maximum included angle between adjacent projection vectors provided in some embodiments of this application;

[0033] Figure 6 This is an architecture diagram of a point cloud-based edge extraction method provided for some embodiments of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the exemplary embodiments of this application clearer, the technical solutions in the exemplary embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described exemplary embodiments are only some embodiments of this application, and not all embodiments.

[0035] Based on the exemplary embodiments shown in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application. Furthermore, although the disclosures in this application are presented by way of one or more exemplary examples, it should be understood that each aspect of these disclosures can constitute a complete technical solution on its own.

[0036] A point cloud is a set of feature points that represent the spatial distribution and surface characteristics of a target within the same spatial reference frame. For example, a point cloud obtained through laser measurement includes three-dimensional coordinates and laser reflection intensity; a point cloud obtained through photogrammetry includes three-dimensional coordinates and color information (RGB). Therefore, a point cloud can be viewed as a three-dimensional feature point image. The attributes of a point cloud include spatial resolution, point accuracy, and surface normal vectors.

[0037] Point cloud edges refer to edge measurement points in a point cloud that can represent the features of an object, used to characterize the object's geometric features. In some embodiments, to accurately extract edge points from a large number of 3D points, depth maps can be used to identify edge points in the point cloud. This involves generating a depth map of the point cloud based on the distance of each point to the camera plane, calculating the height difference between each point based on the depth map, and then extracting points with a height difference greater than a certain threshold as edge points. However, the process of generating the depth map of the point cloud incurs certain losses, affecting the accuracy of the depth map data and leading to a decrease in the precision of edge extraction.

[0038] Therefore, to mitigate the loss problem caused by depth maps, some embodiments may employ a k-dtree algorithm (k-dimensional tree) to perform nearest neighbor search, finding neighboring feature points for each feature point in the point cloud. Projection vectors for each neighboring feature point are then generated, and edge points in the point cloud are identified by the angle between these projection vectors. However, finding neighboring pixels using the k-dtree algorithm consumes a significant amount of search time, leading to decreased efficiency in edge point extraction and negatively impacting the edge extraction performance of the point cloud.

[0039] Based on the above application scenarios, in order to improve the low accuracy of edge extraction, some embodiments of this application provide a point cloud-based edge extraction method, such as... Figure 1 As shown, the procedure includes the following steps:

[0040] S1: Obtain the target point cloud.

[0041] The target point cloud comprises feature points at different coordinates, including both edge and non-edge points, each with its corresponding pixel value. For example, in a point cloud obtained by laser measurement, the pixel value of a feature point is the laser reflection intensity value at that point; in a point cloud obtained by photogrammetry, the pixel value of a feature point is its color information. The target point cloud can be various types of point cloud data, such as point clouds with outer contour boundary points or point clouds with internal folded edge points.

[0042] S2: Detect the neighborhood information of feature points using a voxel nearest neighbor search algorithm.

[0043] After acquiring the target point cloud, it is necessary to use the neighborhood information of feature points to detect the differences between each feature point and its neighboring feature points. Therefore, a voxel nearest neighbor search algorithm is used to obtain the neighborhood information of each feature point in the target point cloud. This neighborhood information refers to the information of the feature points in the neighboring feature points. Using the voxel nearest neighbor search algorithm can reduce the time consumed in obtaining neighborhood information, thereby improving the overall efficiency of edge extraction.

[0044] Therefore, in some embodiments, when detecting the neighborhood information of the feature points using the voxel nearest neighbor search algorithm, a voxel grid is established based on the target point cloud, and the index information of the voxel grid is recorded. That is, the voxel grid needs to divide each feature point of the target point cloud into different grids, and record the grid index information. After recording the index information, the grid index of the feature point is calculated based on the index information, and the neighborhood grid of the feature point is queried through the grid index. The size of the neighborhood grid can be various, such as a 3×3×3 neighborhood grid or a 5×5×5 neighborhood grid. Then, the feature point information of the neighborhood grid is obtained as the neighborhood information of the current feature point.

[0045] For example, such as Figure 2 As shown, Figure 2 Each 3D point in the diagram represents a feature point of the target point cloud. Solid lines represent the voxel mesh, and dashed lines represent the neighborhood mesh. The neighborhood mesh size is 3×3×3. After acquiring the target point cloud, the dimensions of the voxel mesh are set according to the target point cloud. Each point in the target point cloud is then divided into different voxel meshes, and the point indices within each mesh are recorded. The neighborhood mesh of a feature point is then queried based on its point index, and the information of each point within that neighborhood mesh is obtained as the neighborhood information of the current feature point.

[0046] S3: Obtain the differential features of feature points based on neighborhood information.

[0047] After detecting the neighborhood information of a feature point, the difference features of the feature point are obtained according to the neighborhood information to determine whether the feature point is an edge point in the point cloud. The difference features include the smoothed displacement of the feature point and / or the maximum angle between adjacent projection vectors of the feature point. This application provides two methods for obtaining the difference features of feature points, applicable to various edge extraction scenarios. When using the smoothed displacement as the difference feature to determine edge points, curved boundaries and folded edges with large curvature can be detected; when using the maximum angle between adjacent projection vectors as the difference feature to determine edge points, the outer contour boundary and folded edges of the point cloud can be detected.

[0048] like Figure 3As shown, when the difference feature is the smoothed displacement, in some embodiments, when obtaining the difference features of feature points based on neighborhood information, smoothing is performed on the target point cloud according to the neighborhood information to generate a smoothed point cloud. The smoothing process can be performed multiple times to iterate the point cloud data multiple times. Then, the first coordinate of the target feature point in the smoothed point cloud and the second coordinate of the target feature point in the target point cloud are detected. The coordinate difference between the first and second coordinates is calculated as the smoothed displacement. This embodiment uses the smoothed displacement as the difference feature, which is applicable to edge extraction of high-curvature edge points or folded edge points. Curvature refers to the rate of rotation of the tangent direction angle with respect to the arc length at a point on a curve, indicating the degree to which the curve deviates from a straight line. The higher the curvature, the greater the curvature of the curve.

[0049] To improve the accuracy of smoothing, some embodiments employ a bilateral filtering algorithm to smooth the target point cloud. Bilateral filtering is a non-linear filtering method that combines spatial proximity and pixel value similarity in an image. Furthermore, bilateral filtering considers spatial information and grayscale similarity, achieving edge-preserving noise reduction. It is simple, non-iterative, and local.

[0050] For example, such as Figure 4a , 4b As shown, when the edge points of the target point cloud are Figure 4a , 4b When dealing with internal folded edge points or high-curvature edge points, the smoothed displacement can be used as the difference feature of the feature points. First, neighborhood information is obtained using a voxel nearest neighbor search algorithm. Then, bilateral filtering smoothing is performed on each feature point of the target point cloud based on this neighborhood information. Next, multiple bilateral filtering smoothing processes are performed on the smoothed point cloud, such as five iterations. Finally, the displacement of each feature point after multiple iterations of smoothing is recorded as the difference feature of the feature points.

[0051] It should be noted that the smoothing processing method provided in the embodiments of this application can also employ smoothing algorithms such as trilateral filtering, linear interpolation, and MLS (Moving Lest Squares). This application does not impose any limitations on this.

[0052] like Figure 5As shown, when the difference feature is the maximum angle between adjacent projection vectors, in some embodiments, when obtaining the difference feature of feature points based on neighborhood information, plane fitting is performed on the target point cloud according to the neighborhood information, and outliers in the neighborhood feature points are removed by using a preset distance threshold to generate a local fitting plane. That is, plane fitting is performed on the feature points to remove outliers. The feature points are projected onto the local fitting plane to generate the projection vectors of the feature points. Then, the angle between two adjacent projection vectors is calculated, and the maximum value of the angle is recorded as the difference feature of the feature points.

[0053] To facilitate the removal of outliers from the target point cloud, in some embodiments, when removing outliers from neighboring feature points using a preset distance threshold, the system queries neighboring feature points based on neighborhood information and detects the target distance between neighboring feature points and the current feature point. If the target distance is greater than the distance threshold, the neighboring feature points are marked as outliers and removed; if the target distance is less than or equal to the distance threshold, the neighboring feature points are marked as inliers, and a local fitting plane is fitted based on the inliers. By using the distance between neighboring feature points and the current feature point, feature points that are farther away are removed to generate a local fitting plane for the current feature point.

[0054] For example, such as Figure 4c As shown, when the edge points of the target point cloud are Figure 4c When considering the outer contour boundary points, the maximum angle between adjacent projection vectors can be used as the difference feature of the feature points. First, neighborhood information is obtained using a voxel nearest neighbor search algorithm, and then a plane fitting process is performed on the target points in the target point cloud to remove outliers. The target point is the currently processed feature point. A distance threshold k is used to determine whether each neighboring feature point of the target point is an inlier of the local fitting plane. Points less than or equal to the distance threshold k are inliers, and points greater than the distance threshold k are outliers. A uvn coordinate system is established with the target point as the origin. Outliers are removed, and a local fitting plane is fitted based on the inliers. Each feature point and its corresponding inlier are projected onto this plane. The maximum angle between adjacent projection vectors is then recorded as the difference feature of the current feature point. For example... Figure 5 As shown, Figure 5 The included angle θ is the maximum included angle between adjacent projection vectors.

[0055] It is understandable that the two difference features mentioned above are applicable to different types of edge extraction. When it is necessary to detect curved boundaries and folded edges with large curvature, the edge points of the target point cloud are identified by the displacement after smoothing the feature points; when it is necessary to detect the outer contour boundary, the edge points of the target point cloud are identified by the maximum angle between adjacent projection vectors. Therefore, the edge extraction method provided in this application embodiment can be applied to a variety of edge types, thereby improving the universality of edge extraction.

[0056] S4: Calculate the discrimination threshold based on the difference characteristics.

[0057] After obtaining the difference features of each feature point, a set of displacement or maximum angle data can be obtained, and a discrimination threshold for identifying feature points can be calculated using this set of data. In this way, each feature point in the target point cloud can be distinguished based on the difference features, and feature points with large differences can be marked as edge points.

[0058] To facilitate edge point identification, the discrimination threshold can be calculated based on statistical principles. In some embodiments, when calculating the discrimination threshold based on the differential features, the mean and standard deviation of the differential features are obtained, and the discrimination threshold is calculated based on the mean and standard deviation. The standard deviation reflects the dispersion of the differential features in the target point cloud. Combined with the mean of the differential features, the discrimination threshold used to identify edge points can be obtained.

[0059] Therefore, in some embodiments, when calculating the discrimination threshold based on the mean and standard deviation, a threshold factor for the discrimination threshold is obtained, and the product of the threshold factor and the standard deviation is calculated. The sum of this product and the mean is then calculated as the discrimination threshold. The threshold factor is a preset value and can be customized.

[0060] For example, the discrimination threshold can be calculated using the following formula:

[0061] Thresh D =Mean D +m·σ D

[0062] In the formula, Thresh D To determine the threshold, Mean D σ is the average value, m is the preset threshold factor, and σ is the σ value. D The standard deviation is given. When the difference is characterized by displacement, Mean... D With σ D The mean and standard deviation of the displacements; the difference characteristic is the maximum angle between adjacent projection vectors, Mean D With σ D The mean and standard deviation of the maximum included angles.

[0063] S5: Mark feature points with differential characteristics greater than the discrimination threshold as edge points.

[0064] After calculating the discrimination threshold, edge points in the target point cloud can be identified based on this threshold. Each feature point in the target point cloud is detected; if the difference between a feature point and other feature points exceeds the discrimination threshold, it indicates a significant difference between the current feature point and other feature points, and this feature point is marked as an edge point. By using the calculated discrimination threshold to distinguish each feature point in the point cloud, there is no need to convert the point cloud data to other formats, which can mitigate the loss caused by data conversion and improve the accuracy of edge extraction.

[0065] In some embodiments, when feature points with a difference feature greater than a discrimination threshold are marked as edge points, the feature points of the target point cloud are traversed to obtain the difference features of the feature points, i.e., the smoothed displacement of the feature points, and / or the maximum angle between adjacent projection vectors of the feature points. If the difference feature is greater than the discrimination threshold, the feature point is marked as an edge point; if the difference feature is less than or equal to the discrimination threshold, the feature point is marked as a non-edge point, until all feature points in the target point cloud have been traversed to filter out all edge points.

[0066] Based on the above-described point cloud-based edge extraction method, some embodiments of this application also provide a point cloud-based edge extraction device, such as... Figure 6 As shown, the device includes a search module 100, a data processing module 200, and an edge point extraction module 300. Wherein:

[0067] The search module 100 is configured to acquire a target point cloud, which includes feature points with different coordinates; and to detect the neighborhood information of the feature points using a voxel nearest neighbor search algorithm.

[0068] The data processing module 200 is configured to obtain the difference features of the feature points based on the neighborhood information, the difference features including the smoothed displacement of the feature points, and / or the maximum angle between adjacent projection vectors of the feature points; and to calculate a discrimination threshold based on the difference features.

[0069] The edge point extraction module 300 is configured to mark feature points whose difference features are greater than the discrimination threshold as edge points.

[0070] In some embodiments, before obtaining the difference features of the feature points based on the neighborhood information, the data processing module 200 is further configured to generate edge extraction mode selection options, including a smoothing mode option and a neighborhood vector angle mode option. The data processing module can extract the corresponding difference features based on the operation events of the mode selection options. When the smoothing mode option is selected, the data processing module 200 obtains the smoothed displacement as the difference feature; when the neighborhood vector angle mode option is selected, the data processing module 200 obtains the maximum angle between adjacent projection vectors of the feature point as the difference feature.

[0071] As can be seen from the above technical solutions, the edge extraction method and apparatus based on point clouds provided in some embodiments of this application can acquire target point clouds of feature points with different coordinates and detect the neighborhood information of feature points through a voxel nearest neighbor search algorithm. Then, the difference features of feature points are acquired based on the neighborhood information. These difference features include the smoothed displacement of the feature points and / or the maximum angle between adjacent projection vectors of the feature points. A discrimination threshold is calculated based on the difference features, and feature points with difference features greater than the discrimination threshold are marked as edge points. This method can acquire the neighborhood information of each feature point based on the voxel nearest neighbor search algorithm, improving the efficiency of acquiring neighborhood information; and it provides two methods for identifying edge points in the point cloud: smoothing processing or adjacent projection angles. Identifying edge points using the original point cloud data can improve the accuracy of edge extraction, thereby improving the effect of edge extraction.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0073] For ease of explanation, the above description has been provided in conjunction with specific embodiments. However, the above exemplary discussion is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. Various modifications and variations can be obtained based on the above teachings. The selection and description of the above embodiments are for the purpose of better explaining the principles and practical applications, thereby enabling those skilled in the art to better utilize the described embodiments and various different variations of embodiments suitable for specific use considerations.

Claims

1. A point cloud-based edge extraction method, characterized in that, include: Obtain a target point cloud, wherein the target point cloud includes feature points with different coordinates; The neighborhood information of the feature points is detected using a voxel nearest neighbor search algorithm; Based on the neighborhood information, the difference features of the feature points are obtained, and the difference features include the smoothed displacement of the feature points and the maximum angle between adjacent projection vectors of the feature points. Calculate the discrimination threshold based on the aforementioned difference characteristics; Feature points whose difference characteristics are greater than the discrimination threshold are marked as edge points; Calculate the discrimination threshold based on the aforementioned difference features, including: Obtain the mean and standard deviation of the differential features; The discrimination threshold is calculated based on the average value and the standard deviation, including: Obtain the threshold factor of the discrimination threshold; Calculate the product of the threshold factor and the standard deviation; The sum of the product and the average value is calculated and used as the discrimination threshold; Marking feature points with differences greater than the discrimination threshold as edge points includes: Traverse the feature points of the target point cloud to obtain the difference features of the feature points; If the difference feature is greater than the discrimination threshold, the feature point is marked as the edge point; If the difference feature is less than or equal to the discrimination threshold, the feature point is marked as a non-edge point; Obtaining the differential features of the feature points based on the neighborhood information includes: The target point cloud is smoothed according to the neighborhood information to generate a smooth point cloud; Detect the first coordinates of the target feature points in the smoothed point cloud, and detect the second coordinates of the target feature points in the target point cloud; Calculate the coordinate difference between the first coordinate and the second coordinate; The target point cloud is smoothed using a bilateral filtering algorithm.

2. The point cloud-based edge extraction method according to claim 1, characterized in that, The neighborhood information of the feature points is detected using a voxel nearest neighbor search algorithm, including: A voxel grid is established based on the target point cloud, and the index information of the voxel grid is recorded; Calculate the grid index of the feature point based on the index information; The neighborhood grid of the feature point is queried using the grid index; Obtain the feature point information of the neighborhood grid.

3. The point cloud-based edge extraction method according to claim 1, characterized in that, Obtaining the differential features of the feature points based on the neighborhood information includes: The target point cloud is fitted with a plane according to the neighborhood information, and outliers in the neighborhood feature points are removed by a preset distance threshold to generate a local fitting plane. The feature points are projected onto the local fitting plane to generate projection vectors for the feature points; Calculate the angle between two adjacent projection vectors; Record the maximum value of the included angle.

4. The point cloud-based edge extraction method according to claim 3, characterized in that, Outsiders are removed from the neighborhood feature points by using a preset distance threshold, including: Based on the neighborhood information, query the neighboring feature points of the feature point; Detect the target distance between the neighboring feature points and the feature points; If the target distance is greater than the distance threshold, the neighborhood feature points are marked as outliers and the outliers are removed. If the target distance is less than or equal to the distance threshold, the neighborhood information points are marked as inliers, and the local fitting plane is fitted based on the inliers.

5. An edge extraction device based on point clouds, characterized in that, include: The search module is configured to acquire a target point cloud, which includes feature points with different coordinates. The neighborhood information of the feature points is detected using a voxel nearest neighbor search algorithm; The data processing module is configured to obtain the difference features of the feature points based on the neighborhood information, the difference features including the smoothed displacement of the feature points and the maximum angle between adjacent projection vectors of the feature points; and to calculate a discrimination threshold based on the difference features. The edge point extraction module is configured to mark feature points whose difference features are greater than the discrimination threshold as edge points; Calculate the discrimination threshold based on the aforementioned difference features, including: Obtain the mean and standard deviation of the differential features; The discrimination threshold is calculated based on the average value and the standard deviation, including: Obtain the threshold factor of the discrimination threshold; Calculate the product of the threshold factor and the standard deviation; The sum of the product and the average value is calculated and used as the discrimination threshold; Marking feature points with differences greater than the discrimination threshold as edge points includes: Traverse the feature points of the target point cloud to obtain the difference features of the feature points; If the difference feature is greater than the discrimination threshold, the feature point is marked as the edge point; If the difference feature is less than or equal to the discrimination threshold, the feature point is marked as a non-edge point; Obtaining the differential features of the feature points based on the neighborhood information includes: The target point cloud is smoothed according to the neighborhood information to generate a smooth point cloud; Detect the first coordinates of the target feature points in the smoothed point cloud, and detect the second coordinates of the target feature points in the target point cloud; Calculate the coordinate difference between the first coordinate and the second coordinate; The target point cloud is smoothed using a bilateral filtering algorithm.

Images