Research On Scene Point Cloud Segmentation Method Based On Supervoxel

Point cloud segmentation is an important factor in 3D point cloud processing technology,which has a direct impact on the subsequent application of point cloud data.Therefore,the efficient and accurate segmentation of point cloud data is a very important research content in3 D point cloud processing field.Under the above-mentioned background,the 3D point cloud segmentation technology is researched in this paper.Combining the supervoxel algorithm and the concavity and convexity of the 3D point cloud,a scene point cloud segmentation method is proposed,which has fast calculation speed and model advantages such as low complexity.However,this method still has some disadvantages of commonly used algorithms,for instance the confusion of stacked objects and over-segmentation of larger objects.Therefore,this paper combines the supervoxel algorithm and the particle swarm optimization fuzzy clustering algorithm to propose a new three-dimensional point cloud segmentation model,which can effectively avoid stacking objects that are easy to be confused,over-segmented and undersegmented.The specific research contents of this paper are as follows:(1)Based on the super-voxel algorithm,this paper proposes a scene point cloud segmentation method based on the concavity and convexity of the super-voxel.First,octree voxels are used to transform Point clouds,and supervoxels are generated by using curvature,distance and fast point feature histograms(FPFH).Then,based on the region growth algorithm restricted by the concavity and convexity,the initial seed points are selected through the residual value,and the supervoxels are merged according to the concavity and convexity and the similarity distance to realize the scene point cloud division.The OSD-v0.2 data set is used for experimental analysis.The algorithm has a running time of 8.7 seconds,an accuracy rate of0.74,and a recall rate of 0.75.This method can achieve better segmentation results for general regular objects.However,the algorithm still has the drawback of stacking objects and oversegmentation of nonlinear objects,which is not good for segmentation of complex scenes.(2)In order to further improve the segmentation efficiency of point cloud data,this paper proposes a scene based on supervoxel and fuzzy c-means of particle swarm optimization(PFCM)Point cloud segmentation method to solve the defects of stacking objects,oversegmentation with nonlinear objects,etc.The algorithm uses the PFCM model to initially divide the supervoxel,and then divides the supervoxel that are stuck together to complete the scene point cloud segmentation.Tests were performed on the OSD-v0.2 dataset.The accuracy rate of the algorithm in this paper is 0.86,the recall rate is 0.83,and the speed is 7.5 seconds.The segmentation accuracy and recall rate of this algorithm are much higher than the unimproved fuzzy clustering algorithm of particle swarm optimization,increasing by 22% and 16%,respectively.