Research And Application Of Feature Line Extraction Of Point Cloud

With the development of 3d data acquisition technology,people acquire high-precision surface models of objects quickly and effectively in the real world.This greatly promoted the wide application of 3D point cloud model in the field of pattern recognition,3D reconstruction,model segmentation and other fields.In point cloud data processing,feature extraction is the prerequisite and foundation of the follow-up work such as surface reconstruction,which has important practical value in many fields.Therefore,feature extraction has been becoming one of the hotspot of research in the field of point cloud processing.This paper focuses on the extraction of feature lines.For the 3D point cloud model,the feature curve extraction is to analyze and calculate the data points on the surface of the model,identify the feature points therein,and connect them to form the smooth feature curve.This paper studies from three aspects: the extraction of feature points,the fitting of feature lines and the application of feature lines in registration.The main research contents as follows:(1)For the feature point extraction algorithm of scattered point cloud,the traditional method mainly adopts two kinds of methods,feature point extraction based on curvature and feature point extraction based on normal vector.Both of these methods have some problems such as noise sensitivity,high computation time complexity and low efficiency.Therefore,a multi-scale feature point extraction algorithm based on covariance matrix is proposed.By local neighborhood point covariance analysis,to give a measure of the likelihood of the point,then the feature points identified according to the set threshold.Then,depending on the different scales,it carries on regional growth and clustering,and fuses them to obtain the feature points of the model.Adopting the fusion of two scale features can help to make up for the shortcoming of the incomplete extraction of single-scale clustering features,and can better describe the model features.(2)Aiming at the existing feature extraction algorithm,it has the disadvantage of high time cost,low noise tolerance and insensitive to the subtle features of the model.A feature curve fitting algorithm based on covariance matrix and moving least squares is proposed.The previously extracted feature points are grouped into a plurality of ribbon clusters by spatial location distribution,and the key feature points are extracted within each cluster according to the main direction and the direction of the feature line is determined.Then,the key feature points are projected onto the local surface fitted by the moving least squares method,and constantly connected to form a smooth feature line.The experimental results show that compared with the existing feature extraction algorithm,this method has high efficiency,good anti-noise performance,and can obtain smooth feature lines.(3)The feature lines extracted by this algorithm are successfully applied to the registration of 3D model objects.By calculating the differential geometry information of the curve,the 3D model objects are matched.Experiments show that the proposed algorithm can effectively extract the features of 3D model and can be applied well in the registration.