Research On Point Cloud Method Vector Estimation

Terrestrial Laser Scanner(TLS) is a new kind of technology to get surveying and mapping data, which can rapidly obtain high density and high-accuracy point cloud data from surface information of objects. And it has been widely used in reverse engineering, cultural relic protection, digital city, deformation monitoring and some other fields. But there is no topological information in the point cloud acquired by TLS in most time. And the spatial relationship among points need to be constucted by K nearest neighbours. Many data processing are based on consistent oriented normal field, oriented normals can be used to caculate first order approximation of underlying surface and distinguish inside and outside of point cloud. Robust and oriented normal estimation is as difficulty as surface reconstruction. There is inevitable noise and outlier which is caused by defect of instument, mannal misopration and external environment in the measured point data.At present, the main methods of normal estimation of point cloud are based on principal component analysis method, which fitting total least square plane of point and it’s neighbours and using the normal of plane to approximate the normal of point. This algorithm can restrain noise but is sensitive to outlier. In this case, further research should focus on how to estimate exact normal of point cloud when there are outliers in the point cloud.On the basis of existing research results of normal estimation, the further research aiming at how to estimate exact normal of point cloud when there is outlier in the point cloud, the main task in this paperIn the first chapter, the paper introduces state-of-the-art of TLS technology, including laser scanners’operating principles and softwares which are used to process scanning and data. Then introduces the Point Cloud Library(PCL), which is an open source C++ library. At last generalizes the current situation of the emphases and difficulties of normal estimation and orienting.In the second chapter, the paper analyses two kind of K nearest neighbours, introduces the necessity to build the spatial index of scatter point cloud data, analyse K-d tree searching algorithm. And analyses two kind of data format of point cloud. Then K-d tree searching algorithm of PCL has been used to search the neighbours of point cloud.In the third chapter, the paper introduces normal estimation algorithms based on Voroni diagram and local plane fitting. And studies principal component analysis method which is based on local plane fitting, analyses the essence of principal component analysis method, which is total least square. Then the mathematical expression of principal component analysis method has been deduced and principal component analysis method has been used to estimate normal of point cloud.In the fourth chapter, when there are outliers in point cloud, the normal estimation error of principal component analysis method has been studied. In order to eliminate outliers, the closed-form solution of tensor voting has been derived. Then the point has been denoted as ball tensor, and outliers have been eliminated by tensor voting method. The result of eliminate can prove the validity of the method.