Point Cloud Segmentation Based On Adaptive DBSCAN Clustering

With the rapid development of modern science and technology,the wide application of 3D laser scanning technology provides technical support for 3D reconstruction and facilitates the processing of point cloud data.As an important link in the process of 3D reconstruction,point cloud segmentation technology occupies a very important position in target detection,classification,recognition,ground object extraction and other fields,and has different research significance under different application backgrounds.The effect of point cloud segmentation directly affects the quality of 3D reconstruction results,so how to achieve effective point cloud segmentation has always been the focus of research.The main research contents of this paper are as follows:(1)DBSCAN density clustering algorithm: Analyzed and studied the existing DBSCAN clustering algorithm,found any core point in the data set,searched all the data points connected with the core point density,traversed all the core points in the neighborhood of the core point,and looked for the density connected with these points until the last point;the experimental results of point cloud segmentation using traditional methods are used to compare the results of the methods used in the paper.(2)Parameter adaptability: The key to the success of the experiment in the DBSCAN density clustering algorithm is the selection of parameters ? and MinPts.? refers to the neighborhood distance threshold of a sample,that is,the neighborhood radius,MinPts refers to the threshold of the number of samples in the neighborhood where the distance of a sample is.The difference in parameters may lead to completely different results.This paper introduces self-adaptability to the selection of parameters,uses the characteristics of the data set itself to generate candidates?and MinPts parameters,automatically finds the stable interval of the cluster number change of the clustering result,and takes the minimum density threshold in this interval and the corresponding ? and MinPts parameter as optimal parameters.(3)Adaptive DBSCAN clustering algorithm point cloud segmentation: first filter and denoise the point cloud data to filter out those obvious outliers;then adaptively determine the two parameters ? and MinPts and use the kd-tree algorithm for clustering Index the point cloud data one by one,and finally the improved algorithm segmented the outdoor point cloud data and the point cloud data of indoor building facades and corridor walls.It can successfully segment windows,walls,cars,trees and other objects.Experimental results compared with the results of segmentation by traditional methods,the experimental results are verified to prove its feasibility.Experiments show that the adaptive DBSCAN clustering algorithm is superior to the traditional DBSCAN clustering algorithm in terms of segmentation time and segmentation effect.This method provides certain technical support for the application and development of 3D reconstruction technology.The paper has 42 figures,7 tables,and 57 references.