Research And Design Of Point Cloud Recognition Method Combining Local And Global Features

With the continuous deepening of computer vision research and the more convenient way of obtaining 3D data,point cloud recognition technology has become a hot issue in current scientific research,and has played an important role in robot navigation,industrial device sorting,automatic assembly,obstacle detection,biological medicine and cultural relic restoration and other fields.The goal of point cloud recognition is to identify known models in the scene through analysis and matching,which can be mainly divided into two categories: recognition based on local features and recognition based on global features.The local feature-based recognition method shows strong robustness under complex scene,defect or occlusion environment,but the time cost is high.The recognition method based on global features has a very fast matching speed,but the recognition accuracy is not high,and the recognition effect is poor for targets with similar appearance.Based on the above background,this paper takes the steps of point cloud recognition:feature point detection,feature description,feature matching and classification and recognition as the main research object,and completes the following work.(1)as the key point cloud to identify early steps,this paper compared and evaluated the fixed scale feature point detection algorithm of adaptive scale performance,and description of the current mainstream of global and local features in-depth analysis,on the basis of the related experimental results to choose this article to identify feature points extraction and description method of the algorithm.(2)Combining with the existing feature matching strategies,a bidirectional nearest neighbor matching algorithm combining normal information is proposed,which is applied to local feature descriptor matching stage,and good experimental results are obtained.(3)A point cloud recognition algorithm combining local and global features is proposed.The main idea is to use the global feature VFH to quickly identify the KNN of the point cloud target,and select the alternative model from the constructed model base.Secondly,feature matching is carried out between the target and the point cloud of the model base through the local feature SHOT.Finally,Hough voting algorithm is used to strengthen the geometric constraints between the feature points and reduce the mismatching rate.Through the indoor different defect degree of the target object test experiment,compared with the traditional recognition method based on global feature VFH and hoff vote recognition algorithm based on local characteristics,the proposed algorithm obtained good recognition effect,promotive in enhancing cloud recognition rate,on the basis of saving a lot of time overhead,can meet the needs of the point cloud recognition application.