Feature Detective Of Point Cloud

Feature detection of point cloud is the most basic and important link in point cloud data processing.Most algorithms in data processing are related to point cloud feature extraction,such as point cloud registration,segmentation,surface reconstruction,resampling and so on.Among them,point cloud registration technology plays an important role in computer vision,and is widely used in 3D reconstruction,target recognition and tracking,intelligent robot and other fields,which has important research value.The traditional point cloud registration technology based on point features can not guarantee the accuracy and efficiency of the corresponding relationship.With the continuous improvement of the requirements for surveying and mapping results,the requirements for registration technology also increase.Continuing to study point cloud registration technology also plays a positive role in promoting the development of other related fields.Because of the stability of planar features,point cloud registration based on planar features is an important research direction.At present,some scholars have proposed point cloud registration technology based on plane features,but when there is gross error or noise in point cloud data,the registration result may be unsatisfactory.In 3D scene,besides plane features,there are other surface models,such as cylinder.At present,there is relatively little research on cylinder features,and most methods will affect the accuracy of fitting cylinder when the point cloud data is incomplete or sparse.As a continuation of the research on the characteristics of point cloud,this paper also discusses the multivariate time series.For the similarity research of multivariate time series,the current mainstream research method is to promote multivariate time series as unitary time series,which is difficult to balance the contradiction between efficiency and accuracy.Combined with the point cloud,this paper regards the multivariate time series as the result of the arrangement and combination of the point cloud in a special form in space,studies through the characteristics of the point cloud,and extracts the characteristics of similarity.Therefore,this paper mainly makes an in-depth study from three aspects: point cloud plane characteristics,cylindrical characteristics and special point cloud multivariate time series similarity characteristics.The specific contents are as follows:(1)Aiming at the problem of high complexity and slow speed of extracting plane by 3D Hough transform,3D Hough transform is improved,and a point cloud plane registration method based on normal vector is proposed.The plane is extracted by the improved 3D Hough transform,and the plane feature is used instead of the traditional point feature as the registration primitive to establish the plane based coordinate transformation model.Through the analysis of algorithm complexity and experimental verification,it shows that the proposed method improves the speed,does not rely on the initial value,and avoids the trap of local optimization.(2)At present,the method of fitting cylinder will affect the accuracy of fitting cylinder when the point cloud data is incomplete or sparse,and has a certain dependence on the initial value.To solve these problems,a non iterative cylindrical fitting method is proposed in this paper.Through rotation and projection transformation,the three-dimensional problem is transformed into a plane problem,and then the parameters of the cylinder are obtained based on the least square principle.Experiments show that the proposed method is simple in principle,easy to implement,and has good accuracy.The biggest feature is to avoid iterative algorithm and improve the calculation efficiency.(3)The existing research on the similarity of multivariate time series is mainly to improve the similarity of univariate time series.These methods are difficult to balance the contradiction between efficiency and accuracy.This paper regards multivariate time series as a special point cloud,studies the similarity by analyzing the characteristics of multivariate time series,and puts forward a similarity measurement method of multivariate time series based on piecewise linearization.Using the property of spatial polyline,the multivariate time series are projected,and the concepts of filter point,filter line and turning point are introduced.The projection is expressed by piecewise linearization,the pattern representation of multivariate time series is constructed,and the similarity is analyzed by dynamic time bending distance.