A vehicle-mounted laser point cloud change detection method

Abstract

The application relates to a vehicle-mounted laser point cloud change detection method, which comprises the following steps: acquiring two-phase point clouds of the same range, generating octrees corresponding to the two-phase point clouds of the same range respectively; traversing the octree of the latter-phase point cloud, comparing and calculating with the former-phase point cloud, obtaining the Hausdorff distance between the point clouds in each octree box of a first specified level; generating new point cloud scalars, and assigning the Hausdorff distance to each new point cloud scalar one by one; calculating the mean value of the difference scalars in each octree box of a second specified level, obtaining the octree boxes with changes; calculating the center points of each octree box with changes; optimizing the octree boxes with changes through mean shift; manually screening the optimized octree boxes with changes, obtaining the change octree box selection result; and updating the point cloud change database. Compared with the prior art, the application has the advantages of obtaining object-level point cloud changes and the like.

Description

Technical Field

[0001] This invention relates to the field of point cloud change detection, and in particular to a method for detecting point cloud changes using vehicle-mounted laser technology. Background Technology

[0002] Roads form the backbone of a city. However, due to the obstruction of roadside trees, comprehensive and detailed data updates of urban roads are often difficult to achieve through satellite imagery. Instead, updates are typically made quarterly or annually using urban topographic base maps, where vector data is collected and managed uniformly, or localized updates are implemented under project-driven initiatives. Road holographic mapping, as a crucial component of the new basic surveying and mapping system, offers a rapid way to acquire more comprehensive and precise geospatial data through vehicle-mounted laser scanning technology. This meets the demands of high-precision maps and supports the development of intelligent transportation, autonomous driving, and other fields.

[0003] Point cloud data acquired by vehicle-mounted laser scanning is difficult to search and quickly locate due to the lack of topological relationships between points. Furthermore, the reflection points of the same ground object cannot remain consistent across different vehicle laser point cloud scans, making direct change detection challenging. Currently used point cloud change detection methods include object extraction followed by change detection, and calculating the distance between the point cloud and an existing 3D model to identify areas of significant change. However, these methods only detect changes in the point cloud data itself; the scattered distribution of changed point cloud points hinders analysis and processing of the results, and fails to capture the point cloud changes of specific objects. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a vehicle-mounted laser point cloud change detection method for acquiring object-level point cloud changes.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for detecting changes in vehicle-mounted laser point clouds includes the following steps:

[0007] Obtain two point clouds within the same range, the two point clouds including the previous point cloud and the next point cloud, and generate octrees corresponding to the two point clouds within the same range respectively;

[0008] Traverse the octree of the next point cloud and calculate the Hausdorff distance between point clouds within each octree box of the first specified level.

[0009] Based on the original scalar information of the point cloud, new scalar attributes are generated, and the Hausdorff distance is assigned to each new scalar attribute one by one.

[0010] Traverse the octree of the new point cloud scalar, calculate the mean of the difference scalar within each octree box at the second specified level, where the second specified level is the same as the first specified level, and filter the octree boxes based on the mean of the difference scalar and the point cloud density to obtain the octree boxes with changes.

[0011] Iterate through all points in each changing octree box and calculate the center point of each changing octree box;

[0012] Based on the center point of the variable octree box, the variable octree box is optimized by mean offset, and the optimized variable octree box information is stored in a JSON file.

[0013] Obtain the changed octree box selection instructions, filter and optimize the changed octree boxes based on the changed octree box selection instructions, obtain the octree box selection results, and update the changed octree box information in the JSON file according to the box selection results.

[0014] Collect and store the point cloud vector information of the octree selection results, and update the point cloud change database.

[0015] Furthermore, the expression for the mean of the difference scalar within each octree box of the second specified level is:

[0016]

[0017] Where, d m Let dH represent the mean Hausdorff distance in the m-th octree. i Let represent the Hausdorff distance of the i-th point, and n represent the total number of points within the octree box.

[0018] Furthermore, the octree bounding box selection based on the mean of the difference scalar and the point cloud density specifically involves:

[0019] If the mean of the difference scalar is not less than the pre-configured first threshold, then the octree box corresponding to the mean of the difference scalar is selected as the difference octree box.

[0020] Traverse all points in the difference octree bounding box, calculate the point cloud density in the specified neighborhood space of each point in the difference octree bounding box, and if the point cloud density is not less than the pre-configured second threshold, then filter the difference octree bounding box corresponding to the point cloud density as the octree bounding box with change.

[0021] Furthermore, the first threshold is 0.8, and the second threshold is 0.2.

[0022] Furthermore, the expression for the Hausdorff distance between point clouds within each octree box of the first specified level is:

[0023] h(A, B) = max a∈A {min b∈B {d(a, b))}

[0024] Where A and B are the point sets within the octree boxes of the first specified level of the point cloud in the previous period and the point cloud in the next period, respectively, d(a, b) is the Euclidean distance between A and B, and h(A, B) is the Hausdorff distance between A and B.

[0025] Furthermore, the optimization of the varying octree boxes through mean shift specifically involves:

[0026] Calculate the mean offset within a specified region based on the center point of the changing octree box.

[0027] Iteratively calculate the mean offset to obtain the final cluster center based on the mean offset, and determine the optimized and modified octree bounding box based on the cluster center.

[0028] Furthermore, the specific formula for calculating the mean offset is as follows:

[0029]

[0030] Where x is the center point, x i Let x be the i-th sample point, n be the total number of sample points, Sh be a sphere with radius h centered at x, k be the total number of sample points in Sh, and M be the total number of sample points in Sh. h (x) is the offset mean vector of x;

[0031] Furthermore, the specific method for obtaining two point clouds within the same range is as follows: the range of two point clouds in the same region is regularized, and two point clouds within the same range are obtained by cropping and splicing the point clouds.

[0032] Furthermore, the information on the changed octree bounding boxes includes the center point of the changed octree bounding box, the octree level, the octree size, the point cloud name, the detection time, and the name of another point cloud related to the change detection.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] (1) Compared with traditional measurement results, the distribution of changing point clouds is scattered. This invention uses an octree box method to divide the changing point cloud data into various boxes, identify the objects in each box, and obtain the object-level point cloud data changes, which is beneficial for subsequent processing and application of point cloud data.

[0035] (2) For the identified octree boxes, the octree boxes with large changes are selected, and the mean offset is used to further locate the octree boxes, so as to obtain octree boxes with large differences in change and clear objects. At the same time, the octree boxes can be selected by manually inputting the selection command of the octree boxes with changes, so as to obtain flexible, accurate and easy-to-process octree point cloud boxes. Attached Figure Description

[0036] Figure 1 This is a flowchart of the present invention;

[0037] Figure 2 This is a schematic diagram comparing the point clouds from the two phases.

[0038] Figure 3 A schematic diagram of generating an octree from a point cloud;

[0039] Figure 4 A schematic diagram showing the result of assigning values ​​to Hausdorff distance information;

[0040] Figure 5 A schematic diagram of the filtering results for a variable octree box;

[0041] Figure 6 A schematic diagram of the optimized and modified octree bounding box result;

[0042] Figure 7 This is a schematic diagram of the octree selection results. Detailed Implementation

[0043] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0044] This embodiment provides a method for detecting changes in vehicle-mounted laser point clouds, such as... Figure 1 As shown, the steps include:

[0045] S1. First, the point clouds of the previous and subsequent periods within the same region are regularized. Through point cloud cropping and stitching, the consistency of their bounding boxes is ensured, resulting in two point clouds within the same region. The two point clouds within the same region are as follows: Figure 2 As shown.

[0046] S2. Generate octrees for the two point clouds of the same range formed by S1 in two phases.

[0047] S3. For the octrees corresponding to the two point clouds generated in S2, calculate the Hausdorff distance between point clouds within each octree box at the first specified level, and assign the Hausdorff distance to the scalar field of the point cloud. The octrees corresponding to the two point clouds are as follows: Figure 3As shown.

[0048] The specific steps of S3 include:

[0049] S3.1: Traverse the octree of the next point cloud and compare it with the previous point cloud. Calculate the Hausdorff distance within the octree bounding boxes of the two point clouds at the first specified level. This distance is the distance between two subsets in the metric space. It transforms the non-empty subsets of the metric space into the metric space itself. The specific calculation formula is as follows:

[0050] h(A, B) = max a∈A {min b∈B {d(a, b)}}

[0051] Where A and B are the point sets within the octree boxes of the first specified level of the point cloud in the previous period and the point cloud in the next period, respectively, d(a, b) is the Euclidean distance between A and B, and h(A, B) is the Hausdorff distance between A and B.

[0052] This formula can be used to calculate the point cloud distance in units of octree frames.

[0053] S3.2: Based on the existing scalar information of the point cloud, generate new scalar attributes, and assign the Hausdorff distance information calculated in S3.1 to the newly generated scalar attributes of each point. The assignment results of the Hausdorff distance information are as follows: Figure 4 As shown.

[0054] S4. For the first octree level specified in S3, calculate the mean of the octree box difference scalar, filter the octree boxes by the mean of the difference scalar and the point cloud density, obtain the octree boxes with changes, and record the center point coordinates of each octree box with changes.

[0055] The specific steps of S4 include:

[0056] S4.1: Traverse the octree of the new point cloud scalar, and calculate the mean of the difference scalar within each octree box at the second specified level. The second specified level is the same as the first specified level. The specific formula for calculating the mean of the difference scalar is:

[0057]

[0058] Where, d m Let dH represent the mean Hausdorff distance in the m-th octree. i Let represent the Hausdorff distance of the i-th point, and n represent the total number of points within the octree box.

[0059] S4.2: Based on the mean of the difference scalar, the octree box is selected as the difference octree box if the mean of the difference scalar is not less than the pre-configured first threshold.

[0060] In some embodiments, the pre-configured first threshold is 0.8.

[0061] S4.3: Traverse all points within the difference octree bounding box, calculate the point cloud density in the specified neighborhood space of each point within the difference octree bounding box. If the point cloud density is not less than a pre-configured second threshold, then filter the difference octree bounding box corresponding to that point cloud density as a changed octree bounding box. Filtering based on point cloud density can remove octree bounding boxes with generally low point cloud densities, and can also remove point cloud variations related to leaves and branches.

[0062] In some embodiments, the pre-configured second threshold is 0.2.

[0063] S4.4: Traverse all points within the modified octree box, calculate the center point of the modified octree box, and the filtering results of the modified octree box are as follows. Figure 5 As shown.

[0064] S5. Based on the center points of the varying octree bounding boxes, optimize the varying octree bounding boxes using mean offset, and store the optimized varying octree bounding box information in a JSON file. The optimized varying octree bounding box result is as follows: Figure 6 As shown.

[0065] The specific steps of S5 include:

[0066] S5.1: Based on the center point of the varying octree bounding box, calculate the mean offset within a specified area. The specific formula for calculating the mean offset is:

[0067]

[0068] Where x is the center point, x i Let x be the i-th sample point, n be the total number of sample points, Sh be a sphere with radius h centered at x, k be the total number of sample points in Sh, and M be the total number of sample points in Sh. h (x) is the offset mean vector of x.

[0069] The mean offset is calculated iteratively to obtain the final cluster center of the mean offset. Based on the cluster center, the optimized and modified octree boxes are determined, thereby removing the adjacent octree boxes.

[0070] S5.2: Record the center points of the changed octree boxes obtained in S5.1 and store them in a JSON file. The file stores the point cloud name, detection time, the name of another point cloud related to the change detection, and information such as the center of all automatically identified changed boxes, octree level, and octree size.

[0071] S6. Load the point cloud onto the human-computer interaction platform, perform final verification of the bounding box selection results through interaction, and read the relevant JSON-formatted changed bounding box information. Obtain the selection instructions for the changed octree bounding boxes, filter and optimize the changed octree bounding boxes based on the selection instructions, obtain the octree bounding box selection results, and update the changed octree bounding box information in the JSON file according to the selection results. The octree bounding box selection results are as follows: Figure 7 As shown.

[0072] S7. Collect vector information of the point cloud within the octree selection result and store it in the database, and perform local data updates.

[0073] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A vehicle-mounted laser point cloud change detection method, characterized in that, The method comprises the following steps: obtaining two-phase point clouds of the same range, the two-phase point clouds comprising a previous-phase point cloud and a next-phase point cloud, and generating octrees corresponding to the two-phase point clouds of the same range respectively; traversing the octree of the next-phase point cloud, and comparing and calculating the Hausdorff distance between the point clouds in each octree block of a first specified level; based on the original scalar information of the point cloud, generating a new scalar attribute, and assigning the Hausdorff distance to each new scalar attribute; traversing the octree of the new point cloud scalar, and calculating the mean value of the difference scalar in each octree block of a second specified level, the second specified level being the same as the first specified level, and screening the octree block based on the mean value of the difference scalar and the point cloud density to obtain the changed octree block; traversing all points in each changed octree block, and calculating the center point of each changed octree block; based on the center point of the changed octree block, optimizing the changed octree block through mean shift, and storing the optimized changed octree block information in a json file; obtaining a changed octree block selection instruction, screening the optimized changed octree block based on the changed octree block selection instruction to obtain a changed octree block selection result, and updating the changed octree block information in the json file according to the selection result; collecting and storing the point cloud vector information of the changed octree block selection result, and updating the point cloud change database.

2. The vehicle-mounted laser point cloud change detection method according to claim 1, characterized in that, The expression of the mean value of the difference scalar in each octree block of the second specified level is: wherein, denotes the Hausdorff distance mean in the i-th octree, denotes the Hausdorff distance mean in the i-th octree, denotes the Hausdorff distance of the i-th point, denotes the total number of points within the octree box.

3. The vehicle-mounted laser point cloud change detection method according to claim 2, characterized in that, The screening of the octree block based on the mean value of the difference scalar and the point cloud density is specifically: if the mean value of the difference scalar is not less than a first threshold value, the octree block corresponding to the mean value of the difference scalar is screened as a difference octree block; traversing all points in the difference octree block, calculating the point cloud density in the specified neighborhood space of each point in the difference octree block, and if the point cloud density is not less than a second threshold value, the difference octree block corresponding to the point cloud density is screened as a changed octree block.

4. The vehicle-mounted laser point cloud change detection method according to claim 3, characterized in that, The first threshold value is 0.

8.

5. The vehicle-mounted laser point cloud change detection method according to claim 3, characterized in that, The second threshold value is 0.

2.

6. The vehicle-mounted laser point cloud change detection method according to claim 1, wherein, The expression of the Hausdorff distance between the point clouds in each octree block of the first specified level is: wherein, and are the sets of points within the octree boxes of the first specified level of the previous and next point cloud, respectively, is the Euclidean distance between and is the Hausdorff distance between and ​​ 7. The vehicle-mounted laser point cloud change detection method according to claim 1, wherein, The optimization of the changed octree block through mean shift is specifically: calculating the mean shift amount according to the center point of the changed octree block in a specified area, iteratively calculating the mean shift amount to obtain a final clustering center of the mean shift amount, and determining the optimized changed octree block based on the clustering center.

8. The vehicle-mounted laser point cloud change detection method according to claim 7, characterized in that, The specific calculation formula of the mean shift amount is: wherein, is the th sample point, is a sphere of radius h centered at is the total number of sample points in is the offset mean vector of .​ 9. The vehicle-mounted laser point cloud change detection method according to claim 1, wherein, The specific method of obtaining the two-phase point clouds of the same range is to perform range normalization on the two-phase point clouds of the same area, and obtain the two-phase point clouds of the same range through point cloud cropping and splicing.

10. The vehicle-mounted laser point cloud change detection method according to claim 1, wherein, The changed octree block information includes the center point, octree level, octree size, point cloud name, detection time, and another-phase point cloud name related to change detection of the changed octree block.

Images