Research And Design Of Point Cloud Segmentation Algorithm Based On Supervoxel

Point cloud segmentation is a key link in 3D point cloud processing technology.The segmentation results have an important impact on the subsequent 3D reconstruction and target recognition.Accurate and fast segmentation of point cloud data has always been an important research topic in the field of point cloud.On the premise of investigating the research status of point cloud segmentation at home and abroad,aiming at the problems that the existing segmentation algorithms rely on the selection of seed points,fixed voxel resolution,the loss of object boundary details,rough segmentation results and so on,this paper proposes an improved supervoxel point cloud segmentation algorithm,which can effectively retain the object boundary details and realize the accurate segmentation of point cloud data.The main work of this paper is as follows:(1)Segmentation of supervoxels.The point cloud data is segmented,and all points are divided into different regions to obtain uniformly distributed supervoxels attached to the real scene.Firstly,the seed selection problem is transformed into a subset selection problem,and then an optimal selection scheme based on energy descent function is proposed by using local information.The fusion based minimization method is used to generate K supervoxels and K representative points from n points,and then the exchange based minimization method is used to redistribute each non representative point,which corrects the previous wrong allocation points,makes the boundary of supervoxels more smooth and complete,retains the object boundary to a great extent,and improves the segmentation accuracy(2)Polymerization of supervoxels.For the aggregation of the obtained supervoxels,this paper improves the traditional supervoxel aggregation algorithm,no longer uses a single concavity standard,but combines the color information of the point cloud with the geometric measurement,and specifies the rules to transform the two features: the adaptive lambda projects the two distributions onto a Gaussian distribution,and the equalization operation makes δC and δG is evenly distributed over the range.In each iteration,the algorithm combines the areas closest to the comprehensive measurement,and a complete tree view will be generated after the iteration.The time complexity of the algorithm is linearly related to the number of points,and the processing efficiency is high.This paper studies the supervoxel segmentation algorithm,and improves the algorithm in the two parts of supervoxel segmentation and supervoxel aggregation,which further improves the shortcomings of traditional methods.The experimental results show that while ensuring the segmentation speed,the algorithm in this paper is more stable and accurate for the processing of complex scenes,improves the accuracy of object segmentation and boundary division,and has good robustness.