3D Point Cloud Target Recognition Based On Lidar

With the development of computer science and technology and the extensive demand of the society for image recognition,three-dimensional point cloud technology has become a research hotspot in the image field.In many 3D point cloud data acquisition equipment such as Lidar(lidar),Asustek Xtion series and so on,lidar has great advantages,such as high resolution,fast response speed and strong anti-jamming ability,so this paper chooses VLP-16 lidar developed by velodyne company as the acquisition equipment.Because the traditional algorithms have some problems,such as difficulty in recognition and low efficiency in the case of complex scene and occlusion,this paper focuses on the key technologies and algorithms based on point cloud preprocessing and point cloud stitching.3D point cloud target object recognition based on lidar is realized.The specific research contents are as follows:Firstly,the research background and significance of target recognition based on lidar are introduced in detail,and the research status and key research methods of each stage are systematically combed.The main algorithms,main research ideas and key technologies of point cloud target recognition technology are summarized,and the existing problems of the current research technology are analyzed.Secondly,the paper studies the acquisition and imaging principle of lidar,and according to the characteristics of lidar point cloud data,three-dimensional bilateral filtering and statistics-based Gaussian filtering are selected to filter and reduce the noise of the collected data.then,after experimental comparison,the neighborhood-based plane fitting method is selected to remove the ground points,and then the smooth region growth algorithm is used to extract the target point cloud in the sceneThirdly,the paper deeply studies the point cloud stitching technology and algorithm.In order to solve the problem that the existing algorithms can not effectively splice the data collected by lidar into a complete data model,so it is impossible to complete target recognition,this paper divides the point cloud splicing into two parts:coarse splicing and fine splicing,and combines and improves the two algorithms,and proposes an ICP algorithm based on geometric features,which reduces the number of iterations.It also improves the running speed and splicing accuracy of the algorithm.Then,the paper discusses the advantages and disadvantages of the two key point detection algorithms,and carries out several groups of comparative experiments to verify the superiority of the 3D-SIFT algorithm.At the same time,based on the study of the performance of global and local feature descriptors,appropriate descriptors are selected for subsequent target recognition.Finally,combined with the above research content,the paper designs a model base target recognition algorithm based on global and local features,which is divided into two parts:training and recognition.Then the 3D-SIFT algorithm is used to detect the key points and calculate the descriptors,in which the VFH descriptor is used for feature matching for rough recognition,and the FPFH descriptor is registered for precise recognition.At the same time,a large number of experiments and algorithm verification work are carried out based on self-built data sets to verify the accuracy and effectiveness of the algorithm studied in this paper in the case of complex scenes and occlusion.