An Improved Ensembles Algorithm Of Normal Estimation For Point Clouds

As an alternative surface representation,point-based geometry has been drawn increased attention in recent years.Since this method does not have to store or maintain globally consistent topological information,and can provide efficient rendering and flexible geometry processing of highly complex 3D-models,it is more flexible than triangle meshes while handling highly complex or dynamically changing shapes.For point cloud data,the normal,a geometry information,is so important that the accurate estimation of it is a basic step for the operations on point clouds.Given a point cloud that presumably sampled from an unknown surface,the important is how to estimate the normal of each point.Some subsistent algorithms,the fitting surface based algorithm;principal component based algorithm;the standard singular value decomposition based algorithm;the Voronoi based algorithm,for example,give the methods that estimate the normal of point clouds.But a point cloud sampled is usually together with noise which affects the accuracy of estimation of normal.So this kind of algorithm requires strong robust.However,the robust of algorithms mentioned above are not strong so that the estimation of normal is not good.An ensemble normal estimation algorithm based on statistical learning gets a good effect on the data with noise and outliers.But due to the randomicity and the same rate of sample,it is easy to cause non-uniform sample and lose local information,which makes the estimation incorrect.The paper introduces an improved ensembles algorithm.By adding both a sub-block sample strategy and an adaptive sample rate,it covers the shortage of original algorithm.At the same time,the improved algorithm shows a new average formula with weight which enhances the robust of it.