Research On Normal Direction Estimation Algorithm Of Point Cloud Based On Neighborhood Selection

With the popularization and development of 3D laser scanners,the way to obtain 3D point cloud data is becoming more and more convenient.As one of the most important attributes of point cloud data,the normal vector plays an important role in many algorithms.At the same time,high-quality points cloud normal vector can significantly improve the performance of many point cloud data processing algorithms such as surface reconstruction,point cloud segmentation,point cloud denoising and feature extraction.However,due to the limitations of sampling equipment,point cloud data usually has outliers,noise,and non-uniform sampling.Affected by these factors,it is challenging to accurately and quickly estimate the normal vector of sharp feature points.This paper proposes two efficient and accurate point cloud normal estimation algorithms based on neighborhood selection as the core idea: Normal Estimation Based on Difference Accumulation and Subspace Propagation(DASP)and Normal Estimation of Point Cloud Based on Neighborhood Drift(PCND).The DASP algorithm is based on segmentation.The previous segmentation-based normal estimation algorithms mainly construct an incidence matrix based on the normal to achieve neighborhood segmentation.However,in some neighborhoods,neighborhood points with similar initial normals may be located on different smooth surfaces.DASP algorithm not only considers the normal information when calculating the incidence matrix,but also considers the location of the point information.In order to improve the efficiency of the algorithm,the DASP algorithm also proposes the Propagation of subspace structures of normal constraints(SPNC),which uses region growth to achieve neighborhood segmentation.The experimental results show that the algorithm can effectively characterize geometric structures such as smooth surfaces and adjacent surfaces.Although the efficiency and performance of the algorithm have been improved to a certain extent,it is restricted by the shortest path algorithm and has certain limitations in efficiency.Furthermore,This paper proposes a faster algorithm based on the idea of offset neighborhood selection.The core of SPNC algorithm is to select a neighborhood with non-current point as the center point to estimate the normal direction of the current point.The previous algorithm emphasized that the selected neighborhood should contain the current point in the neighborhood selection process.To improve efficiency,the SPNC algorithm relaxes the requirement to: the center point of the selected neighborhood is sufficiently close to the current point.Firstly,the neighborhood of the current point is contructed,and all the neighborhood corresponding to the current point constitutes the candidate neighborhood set.The candidated neighcorhood is evaluated by covariance analysis and the optimal beighborhood is selected for the final normal estimation.The experimental results show that the proposed algorithm is equal to the cutting-edge algorithm in terms of the quality of the normal estimation,and similar to the PCA algorithm in terms of the running speed,which can give consideration to both the normal quality and the computational speed to the greatest extent.