3D LiDAR Point Cloud Feature Extraction Algorithm Based on Distance Coding Adaptation

Abstract

The present invention relates to the field of point cloud feature extraction, and discloses a three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation, which includes the following steps: reading the original unordered point cloud data; calculating the distance information of the point cloud; dividing a frame of point cloud into k rings at equal intervals ΔT, and calculating the ring number corresponding to each point cloud; adaptively calculating and selecting the threshold of feature points according to the coding value; solving the covariance matrix of the local plane to obtain the eigenvector and the corresponding eigenvalue; calculating the dispersion of the local plane; adaptively selecting feature points by comparing the threshold and the dispersion; combining the distance coding information, adaptively selecting different numbers of feature points in different distance range intervals; traversing all the distance coding information values, extracting edge feature points and plane feature points if the conditions are met, and skipping if not. The present invention is applicable to lidars of any type and resolution, and has good robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud feature extraction, and particularly to a three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaption. Background Art

[0002] With the development of technologies such as three-dimensional lidar scanning measurement and structured light, lidar is widely used in fields such as autonomous driving and robotics to obtain surrounding information. The three-dimensional point cloud obtained from the lidar sensor is very dense and disordered. In order to reduce the processing time of the dense point cloud, it is necessary to order the disordered point cloud and extract features, and only extract some points with obvious features to speed up the subsequent point cloud algorithm processing time.

[0003] The existing algorithms for point cloud ordering and then feature extraction mainly include: 1) Calculate which beam of the lidar each point cloud belongs to according to the horizontal angle resolution and vertical angle resolution of the lidar, so as to make the point cloud ordered. Take 10 adjacent points on the same beam to calculate the local curvature, and extract edge feature points and plane feature points according to the size of the curvature. This ordering method depends on the angle resolution of the lidar and has poor robustness for different types of lidars. 2) Project the disordered three-dimensional point cloud onto a two-dimensional image, so that each point cloud corresponds to a pixel in the image, and thus the point cloud becomes ordered. Traverse the pixels on the image and calculate the local point cloud curvature between adjacent pixels, so as to extract the edge feature points and plane feature points of the point cloud. However, the method of projecting the point cloud onto the image makes the three-dimensional point cloud lose one dimension of information. 3) Based on the PCA (Principal Components Analysis) method, find several adjacent point clouds to form a local plane, and extract the edge features and plane features of the point cloud according to the eigenvalues and eigenvectors of the plane. However, this method requires calculating the covariance matrix for each point, and the algorithm is very time-consuming. The characteristic of lidar scanning is that the point cloud obtained in the near area is relatively dense, and the point cloud obtained is sparser as the distance gets farther. The above feature extraction methods can all extract feature points, but a fixed number of feature points are extracted within different scanning distances, and the obtained feature points cannot well reflect the characteristics of lidar scanning, and the point cloud is unevenly distributed in the three spatial dimensions. Therefore, we propose a three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaption. Summary of the Invention

[0004] (1) Technical Problems to be Solved

[0005] In view of the deficiencies of the prior art, the present invention provides a three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation, an algorithm for adaptively extracting lidar point cloud feature points according to coding information, to solve the problem that existing feature extraction algorithms are not robust to multiple types of lidar, and to solve the problem that the feature points extracted by existing algorithms are unevenly distributed in the three-dimensional space and cannot well reflect the characteristics of dense near distances and sparse far distances in lidar scanning.

[0006] (2) Technical solutions

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation, comprising the following steps:

[0009] The first step: Read the original unordered point cloud data;

[0010] The second step: Calculate the distance information of the point cloud;

[0011] The third step: Divide a frame of point cloud into k rings at equal intervals ΔT, and calculate the ring number corresponding to each point cloud in combination with the distance information;

[0012] The fourth step: Traverse the coding values, and adaptively calculate the threshold for selecting feature points according to the coding values;

[0013] The fifth step: Fit the current point and the 5 nearest surrounding points into a local plane, solve the covariance matrix of the local plane, and obtain the eigenvector and the corresponding eigenvalues λ1>λ2>λ3>0;

[0014] The sixth step: Calculate the dispersion of the local plane using the eigenvalues obtained in the fifth step;

[0015] The seventh step: Adaptively select feature points by comparing the threshold and the dispersion;

[0016] The eighth step: Combine the distance coding information to adaptively select different numbers of feature points in different distance range intervals.

[0017] The ninth step: Traverse all the distance coding information values, extract the edge feature points and plane feature points if the conditions are met, and skip if not.

[0018] Preferably, the distance in the second step

[0019] Preferably, the ring number R(i) = D i / ΔT.

[0020] Preferably, the edge point selection threshold in the fourth step Plane point selection threshold

[0021] Preferably, the dispersion degree in the sixth step indicates whether the point represents an edge feature point or a plane feature point. The linearity δ of the edge point is (λ1 - λ2) / λ1, and the planarity ε of the plane point is (λ2 - λ3) / λ1.

[0022] Preferably, the selection basis of the edge points in the seventh step is The selection basis of the plane feature points is

[0023] Preferably, the number of edge feature points in the eighth step is N l = ω l R(i), and the number of feature points of the plane points is N p = ω p R(i).

[0024] Preferably, in the ninth step, when the number of extracted feature points reaches the result calculated in the eighth step, the feature extraction is completed within this distance and then enters the next distance for extraction until the traversal ends.

[0025] (III) Beneficial effects

[0026] Compared with the prior art, the three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation provided by the present invention has the following beneficial effects:

[0027] 1. The three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation uses the distance information of the three-dimensional point cloud to perform ordered coding on the input unordered point cloud. By dividing a frame of point cloud into k rings, the point cloud is mapped to the corresponding ring, and the ring number information of all points is recorded to complete the ordered coding. This method is simple and effective, retains the original information of the point cloud, and is applicable to lidar of any type and resolution, with good robustness. Then, the selection threshold of the feature points is calculated adaptively using the coding information, and the edge feature points and plane feature points are extracted by comparing the relationship between the local plane feature value and the threshold. At the same time, different numbers of feature points are adaptively extracted in different distance ranges according to the distance coding information, so that the extracted feature points conform to the characteristics of the lidar scanning with dense near-distance point cloud and sparse far-distance point cloud, and the distribution of the feature points in the three spatial dimensions is more uniform. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the overall flow of the point cloud feature extraction algorithm of the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] Embodiment

[0031] Please refer to Figure 1 , the three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation provided by the present invention includes the following steps:

[0032] The first step: Read the original unordered point cloud data;

[0033] The second step: Calculate the distance of each point cloud P(x, y, z)

[0034] The third step: Divide a frame of point cloud into k rings with equal intervals ΔT, and calculate the ring number R(i) = D corresponding to each point cloud in combination with the distance information obtained in the second step i / ΔT;

[0035] The fourth step: Traverse the coding values, and adaptively calculate and select the threshold of the feature points according to the coding values obtained in the third step. The threshold for selecting edge points The threshold for selecting plane points

[0036] The fifth step: Fit the current point and the surrounding 5 nearest points into a local plane, solve the covariance matrix of the local plane, and obtain the eigenvector and the corresponding eigenvalues λ1 > λ2 > λ3 > 0;

[0037] The sixth step: Use the eigenvalues obtained in the fifth step to calculate the dispersion of the local plane, and use this dispersion to represent whether the point represents an edge feature point or a plane feature point. The linearity of the edge point δ = (λ1 - λ2) / λ1, and the planarity of the plane point ε = (λ2 - λ3) / λ1;

[0038] The seventh step: Adaptively select feature points by comparing the threshold obtained in the fourth step with the dispersion obtained in the sixth step. The selection basis for edge points is The selection basis for plane feature points is

[0039] The eighth step: Combine the distance coding information obtained in the second and third steps, and adaptively select different numbers of feature points in different distance range intervals. The number of edge feature points is N l = ω l R(i), and the number of feature points of the plane point is Np = ω p R(i);

[0040] Step 9: Traverse all the distance coding information values. If the conditions in Step 7 are met, the edge feature points and plane feature points are successfully extracted; otherwise, skip them. When the number of extracted feature points reaches the result calculated in Step 8, the feature extraction is completed within this distance, and then proceed to the next distance for extraction until the traversal ends.

[0041] Compared with the prior art, the present invention improves the feature extraction algorithm for 3D point clouds, directly using the distance information of the point cloud to perform ordered coding on the unordered point cloud. By dividing a frame of point cloud into multiple rings and mapping the point cloud within the corresponding distance to the ring, the point cloud is numbered, thus completing the ordered coding. Compared with the prior art methods of projecting 3D point clouds onto 2D images for ordering and using the lidar resolution to calculate the point cloud beam number for ordering, our method is simple and efficient, maintaining the original three-dimensional information of the point cloud and having good robustness, which can be applied to various types of lidars. Furthermore, we adaptively calculate the feature point selection threshold according to the distance coding information and dynamically select the number of feature points, so that different numbers of feature points are extracted in different distance intervals. Compared with the existing algorithms for extracting a fixed number of feature points, the adaptive feature point extraction algorithm is more in line with the characteristics of lidar scanning, and the obtained feature points are more evenly distributed in the three spatial dimensions.

[0042] The 3D lidar point cloud feature extraction algorithm based on distance coding adaption provided by the above embodiments of the present invention focuses on using the distance information of the 3D point cloud to perform ordered coding on the input unordered point cloud. By dividing a frame of point cloud into k rings, mapping the corresponding point cloud to the corresponding ring, and recording the ring number information of all points, the ordered coding is completed. This method is simple and effective, retaining the original information of the point cloud and being applicable to lidars of any type and resolution, with good robustness.

[0043] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation, characterized in that, It includes the following steps: The first step: Read the original unordered point cloud data; The second step: Calculate the distance information of the point cloud; The third step: Divide a frame of point cloud into k rings at equal intervals ΔT, and calculate the ring number corresponding to each point cloud in combination with the distance information; The fourth step: Traverse the coding values, and adaptively calculate and select the threshold of the feature points according to the coding values; The fifth step: Fit the current point and the 5 nearest surrounding points into a local plane, solve the covariance matrix of the local plane, and obtain the eigenvector and the corresponding eigenvalues λ1 > λ2 > λ3 > 0; The sixth step: Calculate the dispersion of the local plane using the eigenvalues obtained in the fifth step; The seventh step: Adaptively select feature points by comparing the threshold and the dispersion; The eighth step: Combine the distance coding information, and adaptively select different numbers of feature points in different distance range intervals; The ninth step: Traverse all the distance coding information values, extract the edge feature points and plane feature points if the conditions are met, and skip if not; 2. The 3D lidar point cloud feature extraction algorithm based on distance coding adaptation according to claim 1, wherein: The distance in the second step 3. The 3D lidar point cloud feature extraction algorithm based on distance coding adaptation according to claim 1, wherein: In the third step, the ring number R(i) = D i / ΔT.

4. The three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation according to claim 3, characterized in that: The edge point selection threshold in the fourth step The plane point selection threshold 5. The three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation according to claim 1, wherein: The dispersion in the sixth step indicates whether the point represents an edge feature point or a plane feature point. The linearity δ of the edge point is δ = (λ1 - λ2) / λ1, and the planarity ε of the plane point is ε = (λ2 - λ3) / λ1; 6. The three-dimensional lidar point cloud feature extraction algorithm based on distance coding adaptation according to claim 4, wherein: The selection basis of the edge points in the seventh step is The selection basis of the planar feature points is 7. The 3D lidar point cloud feature extraction algorithm based on distance coding adaptation according to claim 3, characterized in that: In the eighth step, the number of edge feature points is N l = ω l R(i), the number of feature points of the planar points is N p = ω p R(i).

8. The 3D lidar point cloud feature extraction algorithm based on distance coding adaptation according to claim 1, wherein: In the ninth step, when the number of extracted feature points reaches the result calculated in the eighth step, the feature extraction is completed within this distance and proceeds to the next distance for extraction until the traversal ends.

Images