High-efficiency three-dimensional point cloud map construction method

Abstract

The invention discloses a high-efficiency three-dimensional point cloud map construction method, relates to the field of mobile robot SLAM, and solves the problem that it is difficult to analyze local features of cloud data and continuity of laser radar scanning through combination of a dynamic miscellaneous point elimination mechanism and nearest neighbor interpolation. The technical problem that ground miscellaneous points are difficult to remove while point cloud structure features are reserved in a voxel filtering mode is solved, and the method comprises the following steps: extracting an original LEGO LOAM algorithm, and placing extracted ground point clouds in a container; a dynamic miscellaneous point elimination mechanism is introduced, the height difference and density change of the point cloud data and the continuity of laser radar scanning are analyzed, and nearest neighbor interpolation processing is carried out on points in the container; voxel filter plate processing is carried out on the new point cloud in the container after interpolation processing; joining an original LEGO LOAM algorithm architecture, and publishing the ground point cloud and the non-ground point cloud; and reproducing an originally obtained point cloud picture under a KITTI06 public data set, and analyzing and comparing the point cloud picture with the improved L-LEGO LOAM point cloud picture.

Description

Technical Field

[0001] The present invention is applicable to the field of mobile robot SLAM, and specifically relates to a high-efficiency three-dimensional point cloud map construction method. Background Art

[0002] With the rapid development of fields such as autonomous driving, robot navigation, and environmental perception, the accurate construction and efficient processing of three-dimensional environmental information has become one of the key technologies. LiDAR is widely used in these fields because it can provide rich three-dimensional spatial information. Among the many LiDAR three-dimensional point cloud map construction algorithms, the LEGO LOAM lightweight ground-optimized LiDAR odometer and mapping method has attracted much attention due to its lightweight, high efficiency, and ground optimization characteristics. However, the traditional LEGO LOAM algorithm directly relies on the laser lines of multi-line LiDAR to scan the ground during the ground extraction process. This method easily leads to a large number of non-ground clutter in the ground point cloud in complex environments, which in turn affects the accuracy of the map and the accuracy of subsequent navigation and positioning.

[0003] In addition, the huge size and complexity of 3D point cloud data require the algorithm to strike a good balance between computational efficiency and accuracy. Traditional algorithms often have huge computational workloads when processing large amounts of data, making it difficult to process in real time. Especially in highly dynamic environments, the real-time and robustness of the algorithms face huge challenges. At the same time, in order to remove clutter and maintain the integrity of valid point clouds, traditional methods may require complex computational models, which not only increases the computational burden, but may also lead to the loss of valid information.

[0004] Therefore, in order to overcome the shortcomings of the LEGO LOAM algorithm in ground extraction and improve the accuracy and construction efficiency of 3D point cloud maps, it is necessary to optimize and improve the algorithm. Specifically, it is necessary to design an algorithm that can effectively remove ground clutter, retain valid point cloud information, and improve processing speed while ensuring accuracy. In addition, it is also necessary to consider how to better process large-scale point cloud data, reduce computational complexity, and improve the performance of the algorithm in terms of real-time and robustness.

[0005] Based on the above background, the present invention proposes a three-dimensional point cloud map construction method based on the LEGO LOAM algorithm, namely the L-LEGO LOAM algorithm. This algorithm retains the ground extraction module of the original LEGO LOAM algorithm and introduces an efficient dynamic noise removal mechanism and voxel filtering method to significantly improve the extraction accuracy of the ground point cloud and the clarity of the map. At the same time, the ground point cloud is expanded in combination with the nearest neighbor interpolation method to complete the effective point cloud with unclear structure, further improving the integrity and accuracy of the map. In addition, the algorithm also reduces the publishing operation time of the point cloud and improves the real-time performance of map construction by optimizing the processing flow. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a high-efficiency three-dimensional point cloud map construction method for solving the following technical problems: The existing LEGO LOAM algorithm directly relies on the laser lines of multi-line laser radar to scan the ground during the ground extraction process. This method easily leads to a large number of non-ground clutter in the ground point cloud in complex environments, which in turn affects the accuracy of the map and the accuracy of subsequent navigation and positioning. When processing large amounts of data, traditional algorithms often have huge computational complexity and are difficult to process in real time. Especially in highly dynamic environments, the real-time and robustness of the algorithm face huge challenges. At the same time, in order to remove clutter and maintain the integrity of the valid point cloud, traditional methods may require complex computational models, which not only increases the computational burden, but also may lead to the loss of valid information.

[0007] To solve the above problems, the first aspect of the present invention provides a high-efficiency three-dimensional point cloud map construction method, comprising the following steps: S1: Extract the original LEGO LOAM algorithm and put the extracted ground point cloud into a container; S2: Introduce a dynamic noise removal mechanism to analyze the height difference, density change and continuity of the LiDAR scan of the point cloud data, perform nearest neighbor interpolation processing on the points in the container and cover the original container; S3: after interpolation, the point cloud in the new container is subjected to voxel filtering and covered with the previous level container; S4: Connect the original LEGO LOAM algorithm architecture and publish the ground point cloud and non-ground point cloud; S5: Reproduce the original point cloud image under the KITTI_06 public dataset, and analyze and compare it with the improved L-LEGO LOAM point cloud image.

[0008] As a further solution of the present invention, S1 comprises the following steps: The original LEGO LOAM algorithm is used for ground extraction, and the extracted ground point cloud is placed in a container, which contains the point cloud data.

[0009] As a further solution of the present invention, S2 comprises the following steps: A point cloud rough registration algorithm is introduced to remove dynamic points. By comparing the moving distance of the same cluster object in adjacent frames, the point is regarded as a dynamic point and deleted when it is greater than the threshold. The edge point and surface point features are extracted from the segmented point cloud, and feature matching is performed between consecutive frames to find the pose transformation matrix between consecutive frames. The local features of the point cloud data and the continuity of the lidar scan are analyzed. Among them, local features include height differences and density changes; Read the lidar point cloud data from the point cloud container, initialize the data structure, including the marking matrix and label matrix of the ground points, which are used to mark the ground points and assign a label to each point, project the collected point cloud data into a range image, separate the ground points from the non-ground points, and calculate the average value of all points in the separated ground point cloud to obtain the coordinates of the centroid of the ground point cloud as the reference point. By calculating the vertical distance between the two points, the height difference of other ground points relative to the reference point is obtained, and the height difference data is recorded to form a height difference data set; Height difference calculation formula:

[0010] in, For the The height difference of ground points, ( , , ) is the coordinate of the centroid of the ground point cloud, is the total number of ground points, ( , , ) is the The coordinates of ground points; Preprocess the point cloud data, build a KD tree, set the search radius threshold and the minimum point density threshold, and for each point in the point cloud, use the radiusSearch function to search for neighboring points within the radius threshold. The number of neighboring points found is used as the local density value of the point and stored in the vector densities. Create a visited vector to record whether each point has been visited, traverse the density of each point, and skip it if the density of the point is less than the minimum point density threshold. Otherwise, use the breadth-first search algorithm to cluster the point and its neighboring points. Put points that are not less than the minimum density threshold into the same cluster to obtain density clusters that meet the threshold conditions of the point cloud. Calculate the mean, standard deviation, maximum and minimum values ​​of the density clusters, and use the PCLVisualizer visualization class to visualize the clustering results. Collect raw scan data from the LiDAR sensor, perform preprocessing operations on the raw scan data, including denoising and filtering, extract timestamp information from the preprocessed scan data and sort them in chronological order, and for continuous scan data, calculate the average time interval of the entire time series as a temporal continuity indicator; create a PCL, use the VoxelGrid structure in the PCL library to voxelize the point cloud data, set the size of each voxel, call a function to divide the point cloud data into multiple voxels according to the set voxel size, calculate the occupancy in the voxel grid for each pair of consecutive frames, and calculate the overlap ratio as a spatial continuity indicator by comparing the occupancy of the same voxels in the two frames; assign corresponding weights to the temporal continuity indicator and the spatial continuity indicator to obtain a comprehensive continuity evaluation value; Comprehensive continuity evaluation value calculation formula:

[0011] in, is the comprehensive continuity evaluation value, For the The timestamp of the scan, For the The timestamp of the scan, is the number of scans, are the voxels that are occupied simultaneously in both frames, is the total number of voxels in the two frames, is the weight coefficient corresponding to the time continuity index, is the weight coefficient corresponding to the spatial continuity index; Read the lidar point cloud data from the point cloud container, initialize the data structure, including the marking matrix and label matrix of the ground points, project the collected point cloud data into a range image, separate the ground points from the non-ground points, and store the ground points in the point cloud container. Use the breadth-first search algorithm to cluster and segment the point cloud, separate the ground points from the non-ground points, and use the nearest neighbor interpolation method to expand the extracted ground cloud to obtain a new point cloud container.

[0012] As a further solution of the present invention, S3 comprises the following steps: Create a voxel filter object, set the parameters of the voxel filter, perform voxel filtering on the new point cloud container to obtain the filtered point cloud container, add the point cloud in the filtered point cloud container to the original point cloud container for overwriting, and overwrite the previous level container.

[0013] As a further solution of the present invention, S4 comprises the following steps: Connect the original LEGO LOAM algorithm architecture, divide the point cloud data into ground point cloud and non-ground point cloud according to the algorithm processing, create two different ROS publishers for ground point cloud and non-ground point cloud in the point cloud processing code, and use the created publishers to publish the ground point cloud and non-ground point cloud respectively.

[0014] As a further solution of the present invention, S5 comprises the following steps: The original point cloud image is reproduced under the KITTI_06 public dataset, and the original point cloud data is preprocessed and then segmented. The preprocessing operations include filtering and resampling. The processed original point cloud data is visualized and analyzed and compared with the improved L-LEGO LOAM point cloud image.

[0015] Beneficial effects of the present invention: The present invention introduces an efficient dynamic clutter removal mechanism, which can intelligently analyze the local features of point cloud data and the continuity of laser radar scanning, wherein the local features include height differences and density changes, accurately distinguish ground point clouds from non-ground point clouds, effectively remove ground clutter, and significantly improve the accuracy of ground point cloud extraction, laying the foundation for building high-quality three-dimensional maps; The present invention adopts voxel filtering. While retaining the structural features of the point cloud, the present invention further removes outliers and reduces noise interference, thereby improving the accuracy and clarity of the constructed three-dimensional point cloud map. This helps to improve the accuracy and reliability of subsequent navigation positioning, path planning and other applications; The present invention aims to solve the problem of huge 3D point cloud data. The present invention is implemented by optimizing the algorithm and using the nearest neighbor interpolation method with less computational effort to expand the ground point cloud to complete the effective point cloud with unclear structure, while reducing unnecessary computational effort. In addition, by optimizing the overall processing flow, the publishing operation time of the point cloud is reduced, the real-time performance and processing efficiency of the algorithm are improved, so that the 3D map can be quickly constructed on large-scale data sets; The present invention is not only suitable for complex and changeable outdoor environments, but can also effectively cope with interference from external factors such as different lighting conditions and weather changes, thereby improving the robustness and adaptability of the algorithm. This makes the present invention have a wider application prospect in the fields of autonomous driving, robot navigation, etc. The invention provides a new idea and method for the construction of three-dimensional point cloud maps, and promotes the further development and improvement of the technology in this field. Through continuous optimization and improvement, the invention is expected to become one of the important technical means in the field of three-dimensional map construction in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 Design a flow chart for the method of the present invention; Figure 2 This is a comparison chart of the point cloud effect of the algorithm improvement of the present invention; Figure 3 This is a comparison chart of the point cloud publishing efficiency improved by the algorithm of the present invention. Specific implementation methods

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] See also Figure 1-3 As shown, the present invention is a high-efficiency three-dimensional point cloud map construction method, comprising the following steps: S1: Extract the original LEGO LOAM algorithm and put the extracted ground point cloud into a container; S2: Introduce a dynamic noise removal mechanism to analyze the height difference, density change and continuity of the LiDAR scan of the point cloud data, perform nearest neighbor interpolation processing on the points in the container and cover the original container; S3: after interpolation, the point cloud in the new container is subjected to voxel filtering and covered with the previous level container; S4: Connect the original LEGO LOAM algorithm architecture and publish the ground point cloud and non-ground point cloud; S5: Reproduce the original point cloud image under the KITTI_06 public dataset, and analyze and compare it with the improved L-LEGO LOAM point cloud image.

[0020] Specifically, the original LEGO LOAM algorithm is used for ground extraction, and the extracted ground point cloud is placed in a container, which contains point cloud data; the point cloud coarse registration algorithm is introduced to eliminate dynamic points, and the edge point and surface point features are extracted from the segmented point cloud. Feature matching is performed between consecutive frames to find the pose transformation matrix between consecutive frames, and the local features of the point cloud data and the continuity of the lidar scan are analyzed to obtain height difference data, density change data and comprehensive continuity data respectively. The breadth-first search algorithm is used to cluster and segment the point cloud, and the ground points and non-ground points are processed separately. The nearest neighbor interpolation method is used to expand the extracted ground cloud to obtain a new point cloud container; after creating an original point cloud object, a voxel filter object is created, and the parameters of the voxel filter are set to process the original point cloud to obtain the filtered point cloud, and the filtered point cloud is covered in the original point cloud container; the original LEGO LOAM algorithm architecture, divides point cloud data into ground point cloud and non-ground point cloud according to algorithm processing and publishes them; reads point cloud data in KITTI_06 and preprocesses it, uses KITTI object visualization tool for visualization, and analyzes and compares it with the improved L-LEGO LOAM point cloud map.

[0021] In one embodiment of the present invention, the S1 comprises the following steps: The original LEGO LOAM algorithm is used for ground extraction, and the extracted ground point cloud is placed in a container, which contains the point cloud data.

[0022] Specifically, a laser radar sensor is used to collect raw point cloud data, and after denoising the raw point cloud data, the ground segmentation module in the LEGO-LOAM algorithm is used to identify and segment ground points. According to the ground segmentation result, the point cloud data belonging to the ground is extracted, and the extracted ground point cloud is saved in a container.

[0023] In one embodiment of the present invention, the S2 comprises the following steps: A point cloud rough registration algorithm is introduced to remove dynamic points. By comparing the moving distance of the same cluster object in adjacent frames, the point is regarded as a dynamic point and deleted when it is greater than the threshold. The edge point and surface point features are extracted from the segmented point cloud, and feature matching is performed between consecutive frames to find the pose transformation matrix between consecutive frames. The local features of the point cloud data and the continuity of the lidar scan are analyzed. Among them, local features include height differences and density changes; Read the lidar point cloud data from the point cloud container, initialize the data structure, including the marking matrix and label matrix of the ground points, which are used to mark the ground points and assign a label to each point, project the collected point cloud data into a range image, separate the ground points from the non-ground points, and calculate the average value of all points in the separated ground point cloud to obtain the coordinates of the centroid of the ground point cloud as the reference point. By calculating the vertical distance between the two points, the height difference of other ground points relative to the reference point is obtained, and the height difference data is recorded to form a height difference data set; Height difference calculation formula:

[0024] in, For the The height difference of ground points, ( , , ) is the coordinate of the centroid of the ground point cloud, is the total number of ground points, ( , , ) is the The coordinates of ground points; Preprocess the point cloud data, build a KD tree, set the search radius threshold and the minimum point density threshold, and for each point in the point cloud, use the radiusSearch function to search for neighboring points within the radius threshold. The number of neighboring points found is used as the local density value of the point and stored in the vector densities. Create a visited vector to record whether each point has been visited, traverse the density of each point, and skip it if the density of the point is less than the minimum point density threshold. Otherwise, use the breadth-first search algorithm to cluster the point and its neighboring points. Put points that are not less than the minimum density threshold into the same cluster to obtain density clusters that meet the threshold conditions of the point cloud. Calculate the mean, standard deviation, maximum and minimum values ​​of the density clusters, and use the PCLVisualizer visualization class to visualize the clustering results. Collect raw scan data from the LiDAR sensor, perform preprocessing operations on the raw scan data, including denoising and filtering, extract timestamp information from the preprocessed scan data and sort them in chronological order, and for continuous scan data, calculate the average time interval of the entire time series as a temporal continuity indicator; create a PCL, use the VoxelGrid structure in the PCL library to voxelize the point cloud data, set the size of each voxel, call a function to divide the point cloud data into multiple voxels according to the set voxel size, calculate the occupancy in the voxel grid for each pair of consecutive frames, and calculate the overlap ratio as a spatial continuity indicator by comparing the occupancy of the same voxels in the two frames; assign corresponding weights to the temporal continuity indicator and the spatial continuity indicator to obtain a comprehensive continuity evaluation value; Comprehensive continuity evaluation value calculation formula:

[0025] in, is the comprehensive continuity evaluation value, is the timestamp of the first scan, For the The timestamp of the scan, is the number of scans, are the voxels that are occupied simultaneously in both frames, is the total number of voxels in the two frames, is the weight coefficient corresponding to the time continuity index, is the weight coefficient corresponding to the spatial continuity index; Read the lidar point cloud data from the point cloud container, initialize the data structure, including the marking matrix and label matrix of the ground points, project the collected point cloud data into a range image, separate the ground points from the non-ground points, and store the ground points in the point cloud container. Use the breadth-first search algorithm to cluster and segment the point cloud, separate the ground points from the non-ground points, and use the nearest neighbor interpolation method to expand the extracted ground cloud to obtain a new point cloud container.

[0026] Specifically, a point cloud coarse registration algorithm is introduced to identify dynamic points by comparing the moving distances of the same cluster objects in adjacent frames. For each cluster object, its moving distance between adjacent frames is calculated. If the moving distance is greater than a preset threshold, the point is considered to be a dynamic point and is deleted. Edge point and surface point features are extracted from the segmented point cloud, and feature matching is performed between consecutive frames to find the corresponding edge points and surface points. The feature matching results are used to calculate the pose transformation matrix between consecutive frames, and the local features of the point cloud data and the continuity of the lidar scan are analyzed. The local features include height differences and density changes. The lidar point cloud data is read from the point cloud container, the marking matrix and label matrix of the ground points are initialized, the point cloud data is projected into a distance image, and the RANSAC algorithm is used to separate the ground points from the non-ground points. The coordinates of the center of mass are used as the reference point, and the other points are obtained by calculating the vertical distance between the two points. The height difference between the ground point and the reference point is recorded to form a height difference data set; a KD tree is constructed, and a search radius threshold and a minimum point density threshold are set. For each point in the point cloud, neighboring points within the radius threshold are searched to obtain and store the local density value. It is also determined whether each point has been visited. Points that meet the threshold conditions are placed in the same cluster and the results are visualized; the timestamp information is extracted from the preprocessed scan data, and the average time interval is used as the temporal continuity indicator. The point cloud data is voxelized and the size of each voxel is set to 0.1f, 0.1f, 0.1f, indicating that each voxel is a cube with a side length of 0.1 meters. The overlap ratio of the voxels occupied at the same time in the two frames to the total voxels in the two frames is used as the spatial continuity indicator. The temporal continuity indicator and the spatial continuity indicator are combined to obtain a comprehensive continuity evaluation value. and are weight coefficients, is 0.6, is 0.4.

[0027] In one embodiment of the present invention, the S3 comprises the following steps: Create a voxel filter object, set the parameters of the voxel filter, perform voxel filtering on the new point cloud container to obtain the filtered point cloud container, add the point cloud in the filtered point cloud container to the original point cloud container for overwriting, and overwrite the previous level container.

[0028] Specifically, install the PCL point cloud processing library, import related modules, create a voxel filter object after creating an original point cloud object, set the parameters of the voxel filter, the size of each voxel is 0.1f, 0.1f, 0.1f, use the voxel filter to process the original point cloud, obtain the filtered point cloud, and overwrite the filtered point cloud into the original point cloud container.

[0029] In one embodiment of the present invention, the S4 comprises the following steps: Connect the original LEGO LOAM algorithm architecture, divide the point cloud data into ground point cloud and non-ground point cloud according to the algorithm processing, create two different ROS publishers for ground point cloud and non-ground point cloud in the point cloud processing code, and use the created publishers to publish the ground point cloud and non-ground point cloud respectively.

[0030] Specifically, a node is first initialized in ROS for publishing messages, two different ROS publishers are created for ground point clouds and non-ground point clouds, point cloud data is read from the sensor, the ground point cloud and non-ground point cloud are segmented using the RANSAC algorithm, the ground point cloud and non-ground point cloud are extracted according to the segmentation results, the point cloud data is converted into the PointCloud2 message format of ROS, and distributed.

[0031] In one embodiment of the present invention, the S5 comprises the following steps: The original point cloud image is reproduced under the KITTI_06 public dataset, and the original point cloud data is preprocessed and then segmented. The preprocessing operations include filtering and resampling. The processed original point cloud data is visualized and analyzed and compared with the improved L-LEGO LOAM point cloud image.

[0032] Specifically, the point cloud data in KITTI_06 is read, filtered and resampled, and visualized using the KITTI object visualization tool, which is then analyzed and compared with the improved L-LEGO LOAM point cloud image.

[0033] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A high-efficiency three-dimensional point cloud map construction method, characterized in that: The following steps are involved: S1: Extract the original LEGO LOAM algorithm and put the extracted ground point cloud into a container; S2: Introduce a dynamic noise removal mechanism to analyze the height difference, density change and continuity of the LiDAR scan of the point cloud data, perform nearest neighbor interpolation processing on the points in the container and cover the original container; S3: after interpolation, the point cloud in the new container is subjected to voxel filtering and covered with the previous level container; S4: Connect the original LEGO LOAM algorithm architecture and publish the ground point cloud and non-ground point cloud; S5: Reproduce the original point cloud image under the KITTI_06 public dataset, and analyze and compare it with the improved L-LEGO LOAM point cloud image.

2. A high-efficiency three-dimensional point cloud map construction method according to claim 1, characterized in that: The S1 comprises the following steps: The original LEGO LOAM algorithm is used for ground extraction, and the extracted ground point cloud is placed in a container, which contains the point cloud data.

3. The high-efficiency three-dimensional point cloud map construction method according to claim 1, characterized in that: The S2 comprises the following steps: A point cloud rough registration algorithm is introduced to remove dynamic points. By comparing the moving distance of the same cluster object in adjacent frames, the point is regarded as a dynamic point and deleted when it is greater than the threshold. The edge point and surface point features are extracted from the segmented point cloud, and feature matching is performed between consecutive frames to find the pose transformation matrix between consecutive frames. The local features of the point cloud data and the continuity of the lidar scan are analyzed. Among them, local features include height differences and density changes; Read the lidar point cloud data from the point cloud container, initialize the data structure, including the marking matrix and label matrix of the ground points, which are used to mark the ground points and assign a label to each point, project the collected point cloud data into a range image, separate the ground points from the non-ground points, and calculate the average value of all points in the separated ground point cloud to obtain the coordinates of the centroid of the ground point cloud as the reference point. By calculating the vertical distance between the two points, the height difference of other ground points relative to the reference point is obtained, and the height difference data is recorded to form a height difference data set; Height difference calculation formula: in, For the The height difference of ground points, ( , , ) is the coordinate of the centroid of the ground point cloud, is the total number of ground points, ( , , ) is the The coordinates of ground points; Preprocess the point cloud data, build a KD tree, set the search radius threshold and the minimum point density threshold, and for each point in the point cloud, use the radiusSearch function to search for neighboring points within the radius threshold. The number of neighboring points found is used as the local density value of the point and stored in the vector densities. Create a visited vector to record whether each point has been visited, traverse the density of each point, and skip it if the density of the point is less than the minimum point density threshold. Otherwise, use the breadth-first search algorithm to cluster the point and its neighboring points. Put points that are not less than the minimum density threshold into the same cluster to obtain density clusters that meet the threshold conditions of the point cloud. Calculate the mean, standard deviation, maximum and minimum values ​​of the density clusters, and use the PCLVisualizer visualization class to visualize the clustering results. Collect raw scan data from the LiDAR sensor, perform preprocessing operations on the raw scan data, including denoising and filtering, extract timestamp information from the preprocessed scan data and sort them in chronological order, and for continuous scan data, calculate the average time interval of the entire time series as a temporal continuity indicator; create a PCL, use the VoxelGrid structure in the PCL library to voxelize the point cloud data, set the size of each voxel, call a function to divide the point cloud data into multiple voxels according to the set voxel size, calculate the occupancy in the voxel grid for each pair of consecutive frames, and calculate the overlap ratio as a spatial continuity indicator by comparing the occupancy of the same voxels in the two frames; assign corresponding weights to the temporal continuity indicator and the spatial continuity indicator to obtain a comprehensive continuity evaluation value; Comprehensive continuity evaluation value calculation formula: in, is the comprehensive continuity evaluation value, For the The timestamp of the scan, For the The timestamp of the scan, is the number of scans, are the voxels that are occupied simultaneously in both frames, is the total number of voxels in the two frames, is the weight coefficient corresponding to the time continuity index, is the weight coefficient corresponding to the spatial continuity index; Read the lidar point cloud data from the point cloud container, initialize the data structure, including the marking matrix and label matrix of the ground points, project the collected point cloud data into a range image, separate the ground points from the non-ground points, and store the ground points in the point cloud container. Use the breadth-first search algorithm to cluster and segment the point cloud, separate the ground points from the non-ground points, and use the nearest neighbor interpolation method to expand the extracted ground cloud to obtain a new point cloud container.

4. The high-efficiency three-dimensional point cloud map construction method according to claim 1, characterized in that: The S3 comprises the following steps: Create a voxel filter object, set the parameters of the voxel filter, perform voxel filtering on the new point cloud container to obtain the filtered point cloud container, add the point cloud in the filtered point cloud container to the original point cloud container for overwriting, and overwrite the previous level container.

5. The high-efficiency three-dimensional point cloud map construction method according to claim 1, characterized in that: The S4 comprises the following steps: Connect the original LEGO LOAM algorithm architecture, divide the point cloud data into ground point cloud and non-ground point cloud according to the algorithm processing, create two different ROS publishers for ground point cloud and non-ground point cloud in the point cloud processing code, and use the created publishers to publish the ground point cloud and non-ground point cloud respectively.

6. The high-efficiency three-dimensional point cloud map construction method according to claim 1, characterized in that: The S5 comprises the following steps: The original point cloud image is reproduced under the KITTI_06 public dataset, and the original point cloud data is preprocessed and then segmented. The preprocessing operations include filtering and resampling. The processed original point cloud data is visualized and analyzed and compared with the improved L-LEGO LOAM point cloud image.

Images