A Robust Normal Estimation Method for Terrestrial Laser Scanning Point Cloud Based on Minimum Covariance Determinant

A robust normal estimation method based on local plane fitting and minimum covariance determinant (MCD) is proposed for terrestrial laser scanning (TLS) point cloud with gross errors and non-uniform sampling. Firstly, fast library for approximate nearest neighbors algorithm is performed to retrieve k nearest neighbor point set. Then, robust estimation of its covariance is calculated by DetMCD (deterministic MCD) and multivariate Mahalanobis distance. Finally, robust estimation of normal vector is calculated through principal component analysis (PCA) method. Compared with PCA, robust PCA and random sample consensus based normal estimation method, on simulated TLS point cloud, experimental results show that the proposed method can get more accurate normal estimation under the influences of gross errors. And its parallel improvement can meet the requirement of efficiency for large scale TLS point cloud processing. Further experiment on real TLS data from natural terrain shows that the proposed method helps to better Poisson surface reconstruction and prove its effectiveness in practical application.