Extraction And Application Of Point Cloud Feature Line Based On Neural Network

In recent years,with the continuous development of science and technology,the acquisition of3 D data has become faster and more accurate.This also promotes the wide application of 3D point cloud model in 3D reconstruction and other fields.The extraction of point cloud feature line is one of the basic research contents of point cloud data and the basis of subsequent 3D reconstruction.Therefore,point cloud feature line plays an important role in many aspects,and the research on point cloud feature line is also of great significance.This paper focuses on the extraction of feature lines.The process of extracting the feature line of point cloud model is to first identify the feature points through calculation,and then connect the feature points to be called smooth feature line.Because the traditional feature line extraction algorithm has the disadvantages of poor robustness and insensitive to details,this paper proposes a point cloud feature line extraction algorithm based on deep learning,and applies the extracted feature line to point cloud alignment.The main research contents are as follows:(1)According to the practical problems existing in the calculation of the principal curvature direction of scattered point cloud,two methods are given.First,the operation problem of the principal curvature direction is transformed into the problem of finding the extreme value of the local surface method or vector change.Secondly,the pcpnet network is improved,which can be used to calculate the main curvature direction of point cloud.At the same time,the experimental results of the two algorithms are compared,and the better results are obtained for feature line extraction.(2)In view of the shortcomings of the existing feature extraction algorithms that are insensitive to the fine features of the model and poor noise resistance,a calculation method of curvature value and principal curvature direction is proposed based on pcpnet,and a feature extraction algorithm is proposed based on it.The algorithm uses weighted quadratic curve to fit the local curvature distribution,The ridge Valley feature points are identified by determining the distance from the extreme point of the quadratic curve in the direction of maximum principal curvature;Finally,the feature points are connected by establishing the minimum spanning tree(MST)of the refined potential feature points.The experimental results show that compared with the existing feature line extraction algorithms,this method has high efficiency,good noise resistance and can get smooth feature lines.(3)The feature lines extracted by this algorithm are applied to the registration of 3D model objects.Through the feature line extracted by feature detection in this paper,the inflection points on the feature line are calculated,and the 3dpatch calculation descriptor is used to complete the point cloud alignment.The number of inflection points of the feature line is small and the repeatability is high,which not only improves the alignment efficiency,but also ensures the alignment accuracy.