Research Of Point Cloud Segmentation Algorithm Based On Concavity And Convexity In Clutter Scenes

Point cloud segmentation refers to the task of identifying the scene part that corresponds to a single semantic unit.It uses color,depth value,normal vector and other attributes to cluster the point cloud with similar attributes into a connected region to complete the separation of individual objects in the scene.Due to the lack of prior information about the shape and posture of objects and the influence of clutter scenes,point cloud segmentation is still a challenging task.Many robot applications,such as identifying,mastering and manipulating unknown objects,need reliable information about the shape of the target,so that they can perform well.With the continuous development of 3D data acquisition technology,the rapid and effective acquisition of high precision surface model of objects in the real world has become the focus of the field of computer vision.How to segment and extract the point cloud data in cluttered scenes is a technical bottleneck in this field.In the real world,the background is complex and diverse,the objects involved are countless,the size and shape of the material are different,the positions are arranged at will,and there are adjacent,stacked and obscured between each other.Human beings can easily divide every object in the chaotic scene according to the geometry,appearance,color,edge,space position and so on,and the existing visual processing technology is far from the human being in the object segmentation.Therefore,how to accurately and effectively segment objects in chaotic scenes for identification and reconstruction is a more challenging research topic.According to the actual problems above,this paper focuses on point cloud data processing and segmentation,and then discusses the algorithm of chaotic scene point cloud segmentation,and has done the corresponding experiments to verify.The main work of this article is as follows:1.The present situation of the point cloud data processing and segmentation algorithm is analyzed,and the important role of point cloud segmentation in the field of machine vision isbriefly introduced.It is the theoretical basis and technical support for the robot to complete the manipulation of the object in the chaotic scene.2.Introduce the common point cloud segmentation algorithm and introduce the point cloud segmentation algorithm based on the convexity and conciseness,and describe the steps of the segmentation algorithm in detail.This method first decomposes the scene point cloud hyperbody clustering into the adjacency graph based on the voxel grid,then classifies the edges of the adjacency graph to create the convexity graph,and then combines the region growth to merge the convex relations.An unknown object is obtained by the partition.Experimental results show that the proposed method can obtain better segmentation results for general regular objects,but not ideal for the segmentation of objects such as bowls and cups with concave interior.3.In order to further improve the segmentation efficiency of the point cloud data in the above method,a 3D point cloud segmentation algorithm combining the super voxel geometry and color information is proposed on the basis of the concavity and convexity.This algorithm starts from the division of super voxels in the point cloud,Merges the super voxels with a new metric that captures geometric and color information to obtain a 3D segmentation that maintains the hierarchy of the partitioned regions.The algorithm also has a time complexity that is linearly related to the input point cloud size.The experiment compares the improved segmentation algorithm and the common segmentation algorithm from two aspects of the segmentation efficiency and segmentation accuracy,and verifies the effectiveness of the algorithm.