Point Cloud Segmentation Algorithm And Environment Based On Regional Growth Method

With the continuous development of machine vision and image processing technology,point cloud segmentation is the basis of target recognition,point cloud classification,and three-dimensional reconstruction.The segmentation results have an extremely important role in subsequent scene analysis and become a hot research area of modern machine vision.By processing 3D point cloud data,computers can better understand environmental scenes and provide convenience for work and life,such as automatic object detection,scene modeling,and unmanned driving.This article conducts research on point cloud segmentation.By using the point cloud segmentation method based on region growth,the calculation amount of image processing is reduced,and real-time,accurate and effective segmentation of point cloud data is realized.The specific research work in this paper is summarized as follows:In terms of data acquisition,a binocular vision acquisition system is used to obtain stereo image pairs,obtain depth disparity maps,reconstruct 3D scenes,convert the 3D scene data format to obtain 3D point cloud data,and input algorithms for point cloud segmentation.Firstly,the point cloud data is clustered and over-segmented.In three-dimensional space,the clustering and segmentation of point cloud data is based on the clustering algorithm of the original point cloud data collected to reduce the computational complexity,eliminate noise,and improve the accuracy of segmentation.Voxelize the 3D point cloud data,select seed voxels,use clustering algorithms to form super voxels,and then perform plane fitting on the super voxel data to obtain residual values.Select seed super voxels to achieve point cloud data processing.The segmentation process completes clustering over segmentation.Then,segmentation is performed based on the region growing algorithm.The input data is the super voxel obtained through the segmentation step.The region growth is considered in consideration of constraints such as spatial connectivity,surface smoothness,and surface geometric features to ensure the stability of the segmentation algorithm.The segmentation method proposed in this paper avoids the direct processing of point cloud data,improves the robustness of the segmentation algorithm,and ensures a good segmentation boundary.