An Adaptive Method for Rapid Removal of Road Surface Point Cloud

Abstract

The present invention provides an adaptive method for quickly removing road surface point clouds. The method includes: obtaining point cloud data, and based on the point cloud data, extracting bottom point cloud data; constructing a three-dimensional voxel grid for the bottom point cloud data to replace the bottom point cloud data with the centroid points of each grid in the three-dimensional voxel grid, so as to obtain new bottom point cloud data after replacing the bottom point cloud data; based on the RANSAC algorithm, performing plane fitting on the new bottom point cloud data to obtain road surface point clouds, and removing the road surface points corresponding to the road surface point clouds. The present invention is used to solve the defects of poor ground segmentation effect and long time consumption in the prior art in the case of a large-scale scattered point cloud where the proportion of ground points is too small and there are multiple road surface parameters, so as to improve the removal effect, obtain multiple highway parameters, shorten the removal time, and improve the time processing efficiency of ground removal.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an adaptive method for quickly removing road surface point clouds. Background Art

[0002] When using lidar for highway monitoring, the road surface information and target information in the acquired point cloud data are mixed. In order to separate ground points from non-ground points and improve the accuracy of subsequent target detection, it is necessary to extract and remove the highway ground point cloud. Ground filtering technology is a key technology in point cloud processing, which plays an important role in obtaining road surface information, planning driving routes, and improving target detection accuracy.

[0003] Currently, the commonly used ground removal techniques are mainly divided into three categories: grid-based methods, scan line-based methods, and plane fitting-based methods. Grid method: First, the original point cloud is divided into different grid arrays according to certain rules. Then, using the point cloud features in each grid and combining information such as elevation difference, normal vector, and radial gradient between grids, ground points and non-ground points are separated. This method is suitable for removing uneven ground (hillside, jungle ground). Its ground removal effect is closely related to the grid size. If the grid is too small, the computational load is large and it is difficult to ensure real-time performance. If the grid is too large, the ground removal effect is not good. Scan line method: Utilizes the circular scan line of traditional mechanical lidar. By calculating the vertical and horizontal features between adjacent scan lines, the changes in the scan lines are detected to distinguish ground points from obstacles. The scan line method has a relatively fast calculation speed and good ground removal effect for urban scenes, but it depends on the circular scan line and related algorithms cannot be used for the Bernoulli double-twist scan line generated by non-repetitive scanning systems. Plane fitting method uses a plane model to fit a plane from discrete point clouds and optimizes the plane parameters to the maximum extent to achieve the extraction of flat road surfaces (highways, sidewalks).

[0004] Among the existing ground removal techniques, the plane fitting-based method is suitable for highway ground removal. However, by iteratively repeating the point cloud data to fit the mathematical model parameters from a set of samples containing outliers, the time-consuming for road surface removal is long and the time efficiency is low. Summary of the Invention

[0005] The present invention provides an adaptive method for quickly removing road surface point clouds to solve the defects of long time-consuming and low time efficiency in road surface removal in the prior art, realizing reducing the number of iterations, shortening the ground removal time, and improving the time processing efficiency of ground removal.

[0006] The present invention provides an adaptive method for quickly removing road surface point clouds, including:

[0007] Obtaining point cloud data, and adaptively extracting bottom point cloud data based on the point cloud data;

[0008] Construct a three-dimensional voxel grid for the bottom point cloud data, and replace the bottom point cloud data with the centroid points of each grid in the three-dimensional voxel grid to obtain new bottom point cloud data after replacing the bottom point cloud data;

[0009] Based on the RANSAC algorithm, perform plane fitting on the new bottom point cloud data to obtain road surface point clouds, and remove the road surface points corresponding to the road surface point clouds.

[0010] The adaptive road surface point cloud fast removal method provided by the present invention extracts the bottom point cloud data in the point cloud data to increase the proportion of ground points in the point cloud data. Then, it constructs a three-dimensional voxel grid to reduce the data volume of the underlying point cloud data. After that, based on the RANSAC algorithm, it performs plane fitting on the new bottom point cloud data obtained after increasing the proportion of ground points and reducing the data volume. Thus, it can reduce the number of iterations and the data processing volume of the RANSAC algorithm for plane fitting of the point cloud data, increase the proportion of ground points, and reduce the data volume of ground points. Therefore, the number of iterations is reduced, the ground removal time is shortened, and the time processing efficiency of ground removal is improved. Description of the Drawings

[0011] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0012] Figure 1 is one of the flow schematic diagrams of the adaptive road surface point cloud fast removal method provided by the present invention;

[0013] Figure 2 is the schematic diagram of the principle of constructing a three-dimensional voxel grid in the present invention;

[0014] Figure 3 is the schematic diagram of the core idea of the RANSAC algorithm for dividing the ground;

[0015] Figure 4 is the schematic diagram of collecting inclined road surface point clouds in the adaptive road surface point cloud fast removal method provided by the present invention;

[0016] Figure 5 is the schematic diagram of extracting bottom point cloud data in the adaptive road surface point cloud fast removal method provided by the present invention;

[0017] Figure 6 is the original point cloud data map collected in a single highway scene;

[0018] Figure 7It is a schematic diagram of the experimental result of pavement segmentation of the algorithm of the present invention in a single highway scenario;

[0019] Figure 8 It is a schematic diagram of the experimental result of pavement segmentation of the algorithm of the present invention in a single highway scenario;

[0020] Figure 9 It is a schematic diagram of the experimental result of ground segmentation of the RANSAC algorithm in a single highway scenario;

[0021] Figure 10 It is a schematic diagram of the experimental result of ground segmentation of the RANSAC algorithm in a single highway scenario;

[0022] Figure 11 It is a schematic diagram of the experimental result of ground segmentation of the morphological filtering algorithm in a single highway scenario;

[0023] Figure 12 It is a schematic diagram of the experimental result of ground segmentation of the morphological filtering algorithm in a single highway scenario;

[0024] Figure 13 It is a graph of the original point cloud data collected in a multi-highway scenario;

[0025] Figure 14 It is a schematic diagram of the experimental result of ground segmentation of the algorithm of the present invention in a multi-highway scenario;

[0026] Figure 15 It is a schematic diagram of the experimental result of ground segmentation of the algorithm of the present invention in a multi-highway scenario;

[0027] Figure 16 It is a schematic diagram of the experimental result of ground segmentation of the RANSAC algorithm in a multi-highway scenario;

[0028] Figure 17 It is a schematic diagram of the experimental result of ground segmentation of the RANSAC algorithm in a multi-highway scenario;

[0029] Figure 18 It is a schematic diagram of the experimental result of ground segmentation of the morphological filtering algorithm in a multi-highway scenario;

[0030] Figure 19 It is a schematic diagram of the experimental result of ground segmentation of the morphological filtering algorithm in a multi-highway scenario;

[0031] Figure 20 It is a schematic diagram of the structure of the adaptive pavement point cloud rapid removal device provided by the present invention;

[0032] Figure 21 It is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed implementation manners

[0033] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. 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.

[0034] The self-adaptive road surface point cloud rapid removal method of the present invention will be described below with reference to the accompanying drawings of the specification. Please refer to Figure 1 , the present invention provides a self-adaptive road surface point cloud rapid removal method, including:

[0035] Step 10: Obtain point cloud data, and adaptively extract bottom point cloud data based on the point cloud data;

[0036] In this embodiment, the self-adaptive road surface point cloud rapid removal method proposed by the present invention is applied to a highway scene, and point cloud data of the highway scene is collected. The point cloud data collection methods include an image collection method, a lidar collection method, or a combination of the image collection method and the lidar collection method.

[0037] After the point cloud data is collected, the point cloud data is obtained, and the bottom point cloud data in the point cloud data is adaptively extracted. It should be noted that in this step, since the highway scene contains non-highway objects such as woods or railings in addition to the ground, that is, the collected point cloud data contains not only ground points but also non-ground points. Therefore, after the point cloud data of the highway scene is collected, through a certain method, the bottom point cloud data in the point cloud data is adaptively extracted. In this step, the proportion of ground points in the extracted point cloud is significantly increased, the proportion of ground point clouds in the point cloud is increased, and the proportion of non-ground point clouds is reduced. Thus, in the subsequent ground point cloud recognition process, the interference of non-ground points on the road surface point recognition is reduced, and the accuracy of ground recognition and the processing efficiency of ground removal are improved.

[0038] Step 20: Construct a three-dimensional voxel grid for the bottom point cloud data, and replace the bottom point cloud data with the centroid points of each grid in the three-dimensional voxel grid to obtain new bottom point cloud data replacing the bottom point cloud data;

[0039] In this embodiment, please refer to Figure 2, after obtaining the bottom point cloud data, the purpose of constructing a three-dimensional voxel grid for the bottom point cloud data is to construct a three-dimensional voxel grid for the extracted bottom point cloud, and approximate all points in the voxel with the centroid of all points in the voxel. The purpose of this step is to reduce the data volume of the point cloud while retaining the shape characteristics of the road, and improve the subsequent plane fitting speed and road surface removal efficiency. Thus, the centroid points of each grid in the three-dimensional voxel grid are used to replace the bottom point cloud data, so as to replace all points in the voxel grid with the centroid points of the three-dimensional voxel grid, and obtain new bottom point cloud data with reduced data volume of the bottom point cloud data.

[0040] Step 30, based on the RANSAC algorithm, perform plane fitting on the new bottom point cloud data to obtain road surface point cloud, and remove the road surface points corresponding to the road surface point cloud.

[0041] In this embodiment, the RANSAC algorithm is the Random Sample Consensus algorithm. The RANSAC algorithm is a random parameter estimation algorithm. It fits the mathematical model of the road surface from a set of point cloud data containing outliers through an iterative method, so as to remove the road surface points. The RANSAC algorithm has good robustness, but since it identifies the road surface through continuous iteration, the time efficiency is low. In order to improve the credibility of the result, the number of iterations must be increased. Therefore, after increasing the proportion of ground points and reducing the data volume of ground points, based on the RANSAC algorithm, perform plane fitting on the new bottom point cloud data to obtain road surface point cloud, so as to remove the road surface points corresponding to the road surface point cloud, thereby being able to reduce the number of iterations of the RANSAC algorithm, shorten the ground removal time, and improve the time processing efficiency of ground removal.

[0042] Please refer to Figure 3 , the specific process of using the RANSAC algorithm for ground segmentation is as follows:

[0043] (1) Randomly select three non-collinear points , , in the new bottom point cloud data as the minimum sample subset, and calculate the normal vector of the plane where the three points are located:

[0044]

[0045] (2) Calculate the distance from each point in the new bottom point cloud data to this plane:

[0046]

[0047] (3) Set a threshold , and Points are classified as ground points, and the number of ground points is counted.

[0048] (4) Set the number of iterations , repeat steps (1) to (3), and output the ground point with the largest number of points as the final result.

[0049] In the RANSAC algorithm, refers to the maximum number of iterations, and the actual number of iterations is determined during the operation of the algorithm.

[0050]

[0051] Among them, is the confidence level, representing the probability that at least one sampling is an effective sampling in times of sampling in the RANSAC algorithm, represents the probability of randomly drawing inliers, represents the number of data required to calculate the model parameters.

[0052] In this embodiment, by extracting the bottom point cloud data in the point cloud data, the proportion of ground points in the point cloud data is increased. Then, by constructing a three-dimensional voxel grid, the data volume of the bottom point cloud data is reduced. After that, through the RANSAC algorithm, the new bottom point cloud data obtained after increasing the proportion of ground points and reducing the data volume is subjected to plane fitting, so that the number of iterations and the data processing volume of the plane fitting of the point cloud data by the RANSAC algorithm can be reduced, the proportion of ground points is increased, and the data volume of ground points is reduced, thereby reducing the number of iterations, shortening the ground removal time, and improving the time processing efficiency of ground removal.

[0053] In some other embodiments, after step 30, that is, based on the RANSAC algorithm, performing plane fitting on the new bottom point cloud data to obtain road surface point cloud and removing the road surface points corresponding to the road surface point cloud, the following steps are further included:

[0054] Step 40, by traversing the point cloud data, determining the absolute distance from each point in the point cloud data to the plane fitted based on the new bottom point cloud data; Step 50, removing the road surface point cloud with the absolute distance less than the first preset threshold.

[0055] In this embodiment, after removing the ground points through the RANSAC algorithm, the road surface points can also be removed through point cloud spreading. This is because usually the ground is inclined. When the identified ground is inclined, after the above steps of processing, there are still some ground points in the inclined part that cannot be calculated by the RANSAC algorithm. Please refer to Figure 4, during the process of extracting the bottom point cloud, if the ground is inclined, some ground points cannot be collected. Therefore, after the RANSAC algorithm is applied, point cloud spreading is required to remove the entire road surface. The specific process is as follows: By traversing the point cloud, calculate the absolute distance of each point to the fitted plane, and remove the points whose distance is less than the threshold of the points removed.

[0056] In this embodiment, after removing the ground points through the RANSAC algorithm, point cloud spreading is used to find the road surface points of the inclined road surface to be removed, so as to remove the road surface points of the inclined road surface, thereby improving the accuracy of road surface removal.

[0057] In some other embodiments, after step 30, the step of performing plane fitting on the new bottom point cloud data based on the RANSAC algorithm to obtain road surface point cloud and removing the road surface points corresponding to the road surface point cloud, further includes: step 60, recording the initial number of road surface point cloud, and performing loop plane fitting to obtain multiple road parameters, and removing the corresponding multiple road surfaces based on the multiple road parameters; step 70, recording the number of road surface point cloud after each loop when performing the loop; step 80, if the proportion of the number of road surface point cloud in the initial number of road surface point cloud is less than the second preset threshold, then exit the loop.

[0058] In this embodiment, it should be emphasized that this is another inventive point of the adaptive road surface point cloud fast removal method proposed by the present invention, and multiple road surfaces are removed by executing the above loop steps.

[0059] Specifically: Loop execution steps: Based on the RANSAC algorithm, perform plane fitting on the new bottom point cloud data to obtain road surface point cloud, and remove the road surface points corresponding to the road surface point cloud, so as to perform loop plane fitting to obtain multiple road parameters, and thus remove the corresponding multiple road surfaces according to the multiple road parameters. The number of road surface point cloud after each loop needs to be recorded before the loop starts , after that, the number of road surface point cloud extracted in the current loop will be calculated after each loop execution , when the loop ends, exit the loop. Among them, the proportion of the number of road surface point cloud in the initial number of road surface point cloud is expressed as: ; In an alternative embodiment, when When, end the loop and exit the loop step. Among them, the second preset threshold is the loop stop condition, and the second preset threshold is usually set to 50%.

[0060] In some other embodiments, step 10, the step of adaptively extracting the bottom point cloud data based on the point cloud data includes: step 11, based on the point cloud data, determining the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z-axis to obtain the point cloud extraction range of the point cloud data; step 12, based on the point cloud extraction range, adaptively extracting the bottom point cloud data.

[0061] In this embodiment, calculate the point cloud data on axis maximum coordinate value and minimum coordinate value . The purpose of this step is to determine the point cloud extraction range. Further, in a feasible embodiment, after obtaining and , extract axis numerical values within the range of to to obtain the bottom point cloud data. Referring to Figure 5 , the proportion of ground points in the point cloud extracted after this step will be significantly increased. Among them, the point cloud extraction range is not limited in this embodiment and is not limited to , and can be set according to requirements.

[0062] In some other embodiments, step 11, the step of determining the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z-axis based on the point cloud data includes: step 111, traversing at a step size of stepping forward two elements each time; step 112, by comparing the point cloud with a smaller Z-axis coordinate value with the point cloud corresponding to the current minimum coordinate value, and comparing the point cloud with a larger Z-axis coordinate value with the point cloud corresponding to the current maximum coordinate value, determining the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z-axis.

[0063] In this embodiment, in order to obtain and faster and avoid directly comparing each point with the maximum and minimum values, the method shown in the following figure will be used. When traversing the point cloud data, for the index variable , step forward 2 elements each time, compare every two elements as a group, and compare the point with a smaller value with , and compare the point with a larger value with to reduce the number of comparisons.

[0064]

[0065] In this embodiment, by traversing at the step size of two elements, and comparing the point cloud with a smaller Z-axis coordinate value with the point cloud corresponding to the current minimum coordinate value, and comparing the point cloud with a larger Z-axis coordinate value with the point cloud corresponding to the current maximum coordinate value, the maximum and minimum coordinate values of the point cloud data on the Z-axis are obtained, avoiding direct comparison of each point with the maximum and minimum values, reducing the number of comparisons, and further improving the time efficiency of ground removal.

[0066] In some other embodiments, before step 10, the step of acquiring point cloud data and adaptively extracting bottom point cloud data based on the point cloud data, further includes:

[0067] Step 90, acquiring the original point cloud data;

[0068] Step 100, traversing the original point cloud data to determine whether there are invalid points in the original point cloud data;

[0069] Step 110, removing the invalid points in the original point cloud data to obtain the point cloud data.

[0070] In this embodiment, before road surface clearing, invalid points are removed from the point cloud data because the original point cloud data contains some invalid points that need to be removed in advance. Otherwise, it will interfere with the subsequent algorithm processing and reduce the accuracy of road surface point recognition, thereby improving the accuracy and time efficiency of subsequent road surface point recognition.

[0071] In the original point cloud data obtained by lidar, due to being too close or too far away, or due to surface reflection of objects, some invalid points (including points with non-data values or three-dimensional coordinate values all being 0) will be generated. These points will cause difficulties in subsequent processing and must be removed. In the present invention, mainly by traversing the point cloud, it is judged whether there are non-data points or zero points in the original data for removal.

[0072] Furthermore, to verify the correctness and effectiveness of the proposed algorithm, the following respectively conduct segmentation experiments and accuracy evaluations through single-road scenarios and multi-road scenarios, and field-collect the point cloud data in single-road scenarios and multi-road scenarios as the research objects. The experimental platform is an Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz, 8GB random access memory (RAM), 64-bit Windows 10 operating system, Visual Studio 2017 development environment, and the development language is C++. The proposed algorithm is used to conduct ground segmentation experiments on the point clouds of the two scenarios respectively, and compare with the morphological filtering ground segmentation algorithm and the RANSAC ground segmentation algorithm in the PCL open-source point cloud library.

[0073] 1. Single-road scene ground segmentation experiment

[0074] Please refer to Figure 6 , Figure 6 which is the original point cloud data map collected for the single-road scene. The original point cloud data collected for the single-road scene is collected, and the installation height of the lidar is 1.5 m. There is only one flat road in the scene, and at the same time, there are targets such as trees, vehicles, and pedestrians. Among them, the proportion of ground point cloud is about 7.08%. The ground segmentation experiment is carried out using the algorithm in this paper, and the number of iterations is set to 20, and the threshold is set to 0.07. The obtained segmentation results are as shown in Figures 7 - 8 . Figure 7 and Figure 8 are the remaining non-ground point cloud map and the removed ground point cloud map respectively. From the experimental results, it can be known that for this scene, the ground segmentation effect of the present invention is very good, and at the same time, most of the non-ground target points are completely retained.

[0075] To further verify the effectiveness of the method proposed in the present invention, a comparison is made by combining the RANSAC algorithm and the morphological filtering algorithm. For the RANSAC algorithm, the number of iterations is set to 1000, and the threshold is set to 0.07. Please refer to Figures 9 - 10 , Figure 9 and Figure 10 which are the schematic diagrams of the ground segmentation experimental results of the RANSAC algorithm in the single-road scene. Figure 9 and Figure 10 are the remaining non-ground point cloud map and the removed ground point cloud map respectively; for the morphological filtering algorithm, the grid size is set to 0.01, the slope is set to 0.7, and the maximum filtering window is set to 3.00. Please refer to Figures 11 - 12 , Figure 11 and Figure 12 which are the schematic diagrams of the ground segmentation experimental results of the morphological filtering algorithm in the single-road scene. Figure 11 and Figure 12 are the remaining non-ground point cloud map and the removed ground point cloud map respectively. The segmentation effects are as follows: for this scene, the RANSAC algorithm cannot effectively segment the ground because the proportion of ground point cloud in this scene is too small, and the algorithm cannot obtain ground points with high probability during the random sampling stage; the morphological filtering algorithm can segment the ground points, but there is an over-segmentation problem, and some non-ground points are judged as ground points at the junction of the ground and non-ground, resulting in the loss of non-ground points.

[0076] 2. Regarding the ground segmentation experiment of the multi-road scene

[0077] Please refer to Figure 13 , Figure 13The original point cloud data map collected for the multi-road scenario. The original point cloud data collected for the multi-road scenario, with the lidar installed at a height of 7.5 m. There are four roads with different heights and slopes in the scenario, and the road surface has slight undulations. Among them, the proportion of ground point cloud is about 36.25%. The ground segmentation experiment is carried out using the algorithm in this paper, and the number of iterations is set to be 60, and the threshold is 0.07. Please refer to Figures 14 - 15 , Figure 14 and Figure 15 Figure Figure 14 and Figure 15 are respectively the remaining non-ground point cloud map and the removed ground point cloud map. The segmentation result obtained by the method proposed in this invention is as follows: for this scenario, the ground segmentation effect after processing by the method proposed in this invention is good, and non-ground target points such as vehicles, trees, guardrails, and street lamps are completely retained.

[0078] Next, a comparison is made between the RANSAC algorithm and the morphological filtering algorithm. For the RANSAC algorithm, the number of iterations is set to be 800, and the threshold is 0.07. Please refer to Figures 16 - 17 , Figure 16 and Figure 17 Figure Figure 16 and Figure 17 are respectively the remaining non-ground point cloud map and the removed ground point cloud map; for the morphological filtering algorithm, the grid size is set to 0.01, the slope is 0.7, and the maximum filtering window is 3.0. Please refer to Figures 18 - 19 , Figure 18 and Figure 19 Figure Figure 18 and Figure 19 are respectively the remaining non-ground point cloud map and the removed ground point cloud map. The segmentation effects are as follows: for this scenario, the RANSAC algorithm can only effectively segment a part of the ground, and there are a large number of under-segmented points. The reason is that there are multiple road surfaces in this scenario and there are undulations, but the algorithm can only fit one plane model; the morphological filtering algorithm can effectively segment the ground points, but there is still an over-segmentation problem, and some non-ground points are determined as ground points.

[0079] In summary, in this multi-road scenario, the algorithm in this invention can still effectively segment the road surface and can also completely retain the non-ground points. By comparing the RANSAC algorithm and the morphological filtering algorithm, the effectiveness of the algorithm in this paper is verified.

[0080] 3. Precision evaluation

[0081] To quantitatively evaluate the performance of each algorithm and verify the reliability of the algorithm of the present invention, this article will use three indicators: sensitivity ( ), specificity ( ) and algorithm time consumption for quantitative evaluation. and The calculation formulas are as follows:

[0082]

[0083] Among them, is the number of correctly segmented ground points, is the number of non-ground points mis-segmented as ground points, is the actual number of ground points, is the actual number of non-ground points. The larger is, the larger the proportion of correctly segmented ground points, and the better the effect; the larger

[0084]

[0085] Quantitative evaluation results of different algorithms in different scenarios

[0086] is, the larger the proportion of non-ground points mis-segmented as ground points, and the worse the effect. The quantitative evaluation results of different algorithms in different scenarios are shown in the following table. Through the above table, it can be seen that for the actual single-road scene collected, the algorithm of the present invention has the best segmentation effect, with a sensitivity of 100% and a time consumption of only 8 ms; the morphological filtering algorithm can also achieve effective segmentation, but the specificity is relatively high, reaching 4.13%, and the time consumption is as high as 1471 ms, indicating that the proportion of non-ground points mis-segmented as ground points is relatively high and the real-time performance is poor; while for the RANSAC algorithm, due to the small proportion of ground points in the scene, it is unable to effectively collect ground points and has the worst effect. For the actual multi-road scene collected, the RANSAC algorithm has a poor segmentation effect, with a sensitivity of only 48.51%. The reason is that the RANSAC algorithm can only fit one plane model, and applying it to the multi-road scene will cause a large number of ground points to be under-segmented; the morphological filtering algorithm can achieve effective segmentation of the ground, with a sensitivity of up to 95.46%, but at this time the morphological filtering specificity is even higher, reaching 6.96%, and the time consumption is longer, up to 640 ms, indicating that the algorithm mis-segments more non-ground points and has even worse real-time performance. However, the algorithm of the present invention can still ensure a good segmentation effect in the multi-road scene, with a sensitivity of 98.19% and a specificity of 1.04%, and the time consumption is only 57 ms. The reason is that the algorithm of the present invention comprehensively uses a variety of methods to reduce the calculation amount and realizes multiple plane fittings through the way of loop fitting.

[0087] In summary, the algorithm of the present invention has advantages over the RANSAC algorithm and the morphological filtering algorithm in terms of accuracy and efficiency, and has a wider applicable scenario, which verifies the effectiveness of the algorithm in this paper again.

[0088] 4. Summary

[0089] The adaptive method for quickly removing road surface point clouds proposed in the present invention improves the traditional RANSAC algorithm and solves the problem of ground segmentation in large-scale scattered point clouds where the proportion of ground points is too small and there are multiple road surface parameters. Experiments are carried out using the actually collected single-road and multi-road scenarios, and compared with the representative RANSAC algorithm and morphological filtering algorithm in the open-source point cloud library, which proves the advantages of the proposed algorithm in segmentation accuracy and segmentation efficiency.

[0090] Next, the adaptive device for quickly removing road surface point clouds provided by the present invention will be described. The adaptive device for quickly removing road surface point clouds described below can be mutually corresponding and referred to the adaptive method for quickly removing road surface point clouds described above.

[0091] Please refer to Figure 20 , the present invention also provides an adaptive device for quickly removing road surface point clouds, including:

[0092] A point cloud data acquisition module 2010, configured to acquire point cloud data, and adaptively extract bottom point cloud data based on the point cloud data;

[0093] A three-dimensional voxel grid construction module 2020, configured to construct a three-dimensional voxel grid for the bottom point cloud data, so as to replace the bottom point cloud data with the center points of each grid in the three-dimensional voxel grid, and obtain new bottom point cloud data after replacing the bottom point cloud data;

[0094] A plane fitting module 2030, configured to perform plane fitting on the new bottom point cloud data based on the RANSAC algorithm to obtain road surface point clouds, and remove the road surface points corresponding to the road surface point clouds.

[0095] Further, the adaptive device for quickly removing road surface point clouds further includes:

[0096] An absolute distance calculation module, configured to determine the absolute distance from each point in the point cloud data to the plane fitted based on the new bottom point cloud data by traversing the point cloud data;

[0097] A road surface point removal module, configured to remove the road surface point clouds with the absolute distance less than a first preset threshold.

[0098] Further, the adaptive device for quickly removing road surface point clouds further includes:

[0099] The initial road surface point cloud number recording module is used to record the initial road surface point cloud number, perform loop plane fitting, obtain multiple road parameters, and remove the corresponding multiple road surfaces based on the multiple road parameters;

[0100] The road surface point cloud number recording module is used to record the road surface point cloud number after each loop when performing the loop;

[0101] The loop exit module is used to exit the loop if the proportion of the road surface point cloud number to the initial road surface point cloud number is less than the second preset threshold.

[0102] Furthermore, the point cloud data acquisition module is further used for:

[0103] Based on the point cloud data, determine the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z axis, and obtain the point cloud extraction range of the point cloud data;

[0104] Based on the point cloud extraction range, adaptively extract the bottom point cloud data.

[0105] Furthermore, the point cloud data acquisition module is further used for:

[0106] Traverse at a step size of stepping forward two elements each time;

[0107] By comparing the point cloud with a smaller Z-axis coordinate value with the point cloud corresponding to the current minimum coordinate value, and comparing the point cloud with a larger Z-axis coordinate value with the point cloud corresponding to the current maximum coordinate value, determine the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z axis.

[0108] Furthermore, the adaptive road surface point cloud fast removal device further includes:

[0109] The original point cloud data acquisition module is used to acquire the original point cloud data;

[0110] The invalid point detection module is used to traverse the original point cloud data to determine whether there are invalid points in the original point cloud data;

[0111] The invalid point removal module is used to remove the invalid points in the original point cloud data to obtain the point cloud data.

[0112] Figure 21 Illustrated is a schematic diagram of the physical structure of an electronic device, such as Figure 21As shown, the electronic device may include: a processor 2110, a communications interface 2120, a memory 2130, and a communication bus 2140. Among them, the processor 2110, the communications interface 2120, and the memory 2130 communicate with each other through the communication bus 2140. The processor 2110 can call the logical instructions in the memory 2130 to execute an adaptive road surface point cloud fast removal method, which includes: obtaining point cloud data, and adaptively extracting bottom point cloud data based on the point cloud data; constructing a three-dimensional voxel grid for the bottom point cloud data to replace the bottom point cloud data with the centroid points of each grid in the three-dimensional voxel grid, obtaining new bottom point cloud data after replacing the bottom point cloud data; based on the RANSAC algorithm, performing plane fitting on the new bottom point cloud data to obtain road surface point cloud, and removing the road surface points corresponding to the road surface point cloud.

[0113] In addition, when the logical instructions in the above-mentioned memory 2130 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0114] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the adaptive road surface point cloud fast removal method provided by the above-mentioned various methods. The method includes: obtaining point cloud data, and adaptively extracting bottom point cloud data based on the point cloud data; constructing a three-dimensional voxel grid for the bottom point cloud data to replace the bottom point cloud data with the centroid points of each grid in the three-dimensional voxel grid, obtaining new bottom point cloud data after replacing the bottom point cloud data; based on the RANSAC algorithm, performing plane fitting on the new bottom point cloud data to obtain road surface point cloud, and removing the road surface points corresponding to the road surface point cloud.

[0115] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the adaptive road surface point cloud fast removal method provided by the above-mentioned various methods. The method includes: obtaining point cloud data, and adaptively extracting bottom point cloud data based on the point cloud data; constructing a three-dimensional voxel grid for the bottom point cloud data, so as to replace the bottom point cloud data with the center-of-gravity points of each grid in the three-dimensional voxel grid, and obtaining new bottom point cloud data after replacing the bottom point cloud data; based on the RANSAC algorithm, performing plane fitting on the new bottom point cloud data to obtain road surface point cloud, and removing the road surface points corresponding to the road surface point cloud.

[0116] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0117] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An adaptive method for quickly removing road surface point clouds, characterized in that, it includes: Obtain point cloud data, and adaptively extract bottom point cloud data based on the point cloud data; Construct a three-dimensional voxel grid for the bottom point cloud data, and use the centroid points of each grid in the three-dimensional voxel grid to replace the bottom point cloud data to obtain new bottom point cloud data after replacing the bottom point cloud data; Based on the RANSAC algorithm, perform plane fitting on the new bottom point cloud data to obtain road surface point clouds; by traversing the point cloud data, determine the absolute distance from each point in the point cloud data to the plane fitted based on the new bottom point cloud data; remove the road surface point clouds with the absolute distance less than the first preset threshold; record the initial number of road surface point clouds, and perform loop plane fitting to obtain multiple road parameters, and remove the corresponding multiple road surfaces based on the multiple road parameters; record the number of road surface point clouds after each loop when performing the loop; if the proportion of the number of road surface point clouds in the initial number of road surface point clouds is less than the second preset threshold, exit the loop.

2. The adaptive method for quickly removing road surface point clouds according to claim 1, characterized in that, the step of adaptively extracting bottom point cloud data based on the point cloud data includes: Based on the point cloud data, determine the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z axis to obtain the point cloud extraction range of the point cloud data; Based on the point cloud extraction range, adaptively extract bottom point cloud data.

3. The adaptive method for quickly removing road surface point clouds according to claim 2, characterized in that, the step of determining the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z axis based on the point cloud data includes: Traverse according to a step size of stepping forward two elements each time; By comparing the point cloud with a smaller Z-axis coordinate value with the point cloud corresponding to the current minimum coordinate value, and comparing the point cloud with a larger Z-axis coordinate value with the point cloud corresponding to the current maximum coordinate value, determine the maximum coordinate value and the minimum coordinate value of the point cloud data on the Z axis.

4. The adaptive method for quickly removing road surface point clouds according to any one of claims 1 to 3, characterized in that, before the step of obtaining point cloud data and adaptively extracting bottom point cloud data based on the point cloud data, it further includes: Obtain the original point cloud data; Traverse the original point cloud data to determine whether there are invalid points in the original point cloud data; Remove the invalid points in the original point cloud data to obtain the point cloud data.

5. An adaptive device for quickly removing road surface point clouds, characterized in that, it includes: A point cloud data acquisition module for obtaining point cloud data and adaptively extracting bottom point cloud data based on the point cloud data; A three-dimensional voxel grid construction module for constructing a three-dimensional voxel grid for the bottom point cloud data, and using the centroid points of each grid in the three-dimensional voxel grid to replace the bottom point cloud data to obtain new bottom point cloud data after replacing the bottom point cloud data; A plane fitting module is used to perform plane fitting on the new bottom point cloud data based on the RANSAC algorithm to obtain road surface point clouds; by traversing the point cloud data, determine the absolute distance from each point in the point cloud data to the plane fitted based on the new bottom point cloud data; remove the road surface point clouds with the absolute distance less than a first preset threshold; record the initial number of road surface point clouds, and perform loop plane fitting to obtain multiple road parameters, and remove the corresponding multiple road surfaces based on the multiple road parameters; record the number of road surface point clouds after each loop when performing the loop; if the proportion of the number of road surface point clouds in the initial number of road surface point clouds is less than a second preset threshold, exit the loop.

6. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the steps of the adaptive road surface point cloud fast removal method according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium, on which a computer program is stored, characterized in that, when the computer program is executed by a processor, it implements the steps of the adaptive road surface point cloud fast removal method according to any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps of the adaptive road surface point cloud fast removal method according to any one of claims 1 to 4.

Images