Research On Unmanned Vehicle Point Cloud Clustering Based On Improved DBSCAN Algorithm

Point cloud clustering refers to constructing a point cloud map through lidar and clustering the discrete points in the map into various wholes.It is one of the environmental perception detection technologies for unmanned vehicles.Classical point cloud clustering algorithms mainly include partition-based clustering algorithm,hierarchical-based clustering algorithm,densitybased clustering algorithm,grid-based clustering algorithm,distance-based clustering algorithm and hybrid clustering algorithm.Density-based clustering algorithms are widely used,mainly including Ordering Points To Identify the Clustering Structure algorithm and Density-Based Spatial Clustering of Applications with Noise algorithm.The advantage of Density-Based Spatial Clustering of Applications with Noise algorithm is that it can find point clouds of any shape.Based on the analysis and research of Density-Based Spatial Clustering of Applications with Noise algorithm,aiming at some of the shortcomings of Density-Based Spatial Clustering of Applications with Noise algorithm,this paper improves Density-Based Spatial Clustering of Applications with Noise algorithm by using K-Dimensional-Tree and standardized European distance to realize the point cloud clustering optimization of unmanned vehicles.In terms of point cloud preprocessing,t the point cloud is beneficial to subsequent processing by transforming the original point cloud format,coordinate transformation and reflection intensity compensation.The amount of point cloud data is large and there are outliers.By comprehensively utilizing the advantages of statistics outlier removal algorithm and voxelgrid algorithm,the hybrid filtering method is used to greatly reduce the amount of point cloud data and remove outliers.In the aspect of point cloud segmentation,the ground point cloud is cleared by the point cloud segmentation method.The preprocessed point cloud data contains ground point cloud and nonground point cloud,and the existence of ground point cloud will interfere with obstacle clustering.The Random Sample Consensus method is selected for point cloud segmentation and compared with the minimum shear segmentation algorithm.According to the comparison of the simulation results,the random sampling consistency method has better robustness.Clouds work better and segmentation is processed faster.In point cloud clustering,in view of the shortcomings of Density-Based Spatial Clustering of Applications with Noise algorithm,K-Dimensional-Tree method and standardized Euclidean distance method are adopted.The K-Dimensional-Tree method is added to make the algorithm search faster,reduce the time for the algorithm to search in the neighborhood,and improve the efficiency of the algorithm;The standardized Euclidean distance is used to replace the original Euclidean distance,so that the algorithm calculates the distance between each point.The distance value is more accurate,the clustering effect of the obstacle point cloud is improved,and the accuracy of the algorithm is improved.By building Ubuntu system and PCL simulation environment under VMware WorkstationPro15.5.0,a series of simulations were performed on the processed point cloud data,and the improved algorithm was compared with Density-Based Spatial Clustering of Applications with Noise,Euclidean clustering,regional growth clustering and other algorithms.The clustered point cloud satisfies the obstacle clustering effect,which verifies the accuracy of the improved algorithm.