Object Recognition In Point Cloud Of Single-view Scene

In recent years,the 3D scanning technology has developed rapidly,and the acquisition of point cloud data has become convenient and fast.Because point clouds have object depth information,more and more scholars have begun to focus on using point clouds for object recognition.After summarizing and researching the current status of research at home and abroad,the paper proposes two recognition algorithms for fixed and non-fixed scenes respectively,and proposes improvements in point cloud filtering,normal calculation,key point extraction,feature description,and feature matching.A large number of comparative experiments with commonly used recognition algorithms verify the accuracy and robustness of the recognition algorithm in this paper.This paper first introduces the research background,significance,and common applications of the subject,and describes in detail the development status,technical difficulties,and application prospects of point cloud recognition technology.This paper introduces the point cloud representation method,acquisition,preprocessing,and so on.This paper introduces several stages of the model-based point cloud recognition process: finding key points,feature descriptors,feature matching,and solving transformation matrices.For point cloud recognition in non-fixed scenes,this paper proposes a point cloud recognition algorithm based on voxel shape descriptors.This paper improves the normal estimation algorithm,finds the key points in the point cloud by using the neighborhood normal information,uses the voxel shape descriptor to describe the features,and uses an improved Huff voting algorithm to identify objects in the scene point cloud.While using local features,it effectively fuses global information and improves the speed and accuracy of the recognition algorithm.Finally,the recall rate was improved by a hypothesis verification algorithm based on simulated annealing.For the point cloud recognition in fixed scenes,this paper proposes a point cloud recognition algorithm based on local surface feature histograms.This paper uses cyclic voxel filtering to filter point clouds to a specified resolution,clusters scene point clouds,uses local mean curvature to find key points,uses local surface feature histograms to describe key points,and uses positional relationships between key points for feature matching,judge the recognition result according to the overlap rate.Due to the use of block recognition,the difficulty of recognition is greatly reduced,and it has higher accuracy and robustness.Finally,the algorithm of this paper is implemented in the VS2017 environment,and the accuracy and robustness of the point cloud recognition algorithm in this paper is verified by comparing experiments with common point cloud recognition algorithms.