The Research On Object Classification Based On 3D Point Cloud Covariance Descriptors And Point Cloud Segmentation Technology

Real-time perception of its surroundings and real-time processing of information from the surrounding environment are the most important functions of mobile robot.Compared to a simple planar image,3D point cloud data not only contains the color information of the object,but also contains more geometric information.Image recognition methods based on covariance descriptors have been developed for decades,but how to effectively apply covariance descriptors to 3D point cloud object recognition and classification is a research hotspot in recent years.This paper studies the application of local descriptors,global descriptions in object recognition and classification in 3D point clouds and the mismatch correction algorithm in the point cloud matching process.At the same time,aiming at the characteristics of non-uniformity,disorder,and sparsity of point clouds,this paper proposes a 3D point cloud segmentation method based on topological structure invariance.The main innovative research results of this paper are as follows:(1)Aiming at the problems of large dimension for general descriptors,sensitivity to noise,and long matching operation time,a new covariance descriptor is proposed in conjunction with the 3D point cloud.Combining with the characteristics of point cloud covariance feature descriptor,a corresponding mismatch correction algorithm is proposed,which effectively reduces the number of mismatch in the process of descriptor matching,Therefore,the accuracy and robustness of object recognition is improved.(2)Combining the characteristics of fast processing speed of global descriptors,feature descriptors are extracted from geometric information of 3D point cloud's color,depth,shape,size and other aspects.Then,serial fusion is carried out to generate the global feature descriptors of different combinations.Through the classification experiments of instance and category,the most effective feature vector combination in the object classification is verified and obtained,(3)Since the 3D point cloud has the characteristics of incomplete,sparsity,disorder,dispersion and so on,the idea of topology invariance is introduced into 3D point cloud segmentation based on the topological persistence of computing point cloud datasets at different spatial resolutions and then a new point cloud segmentation algorithm is proposed.Through experimental verification,this algorithm can effectively deal with the point cloud segmentation under the interference conditions such as occlusion and noise.