Parameter-Free Outlier Removal of 3D Point Clouds with Large-Scale Noises

3D point clouds derived from either multi-viewbased techniques or direct laser scanners are inevitably contaminated with severe outliers. This paper presents LSNOR, a parameter-free density-based Outlier Removal approach for point clouds corrupted by Large-Scale Noises. The main contributions are three-fold. (i) A local consistency factor (LCF) is proposed to indicate the local density similarity of points. Based on LCF, parameter estimation and cluster screening are performed via consistency checking. In particular, unlike most of the density-based methods requiring user interactions for parameter determination, the proposed approach realizes automated parameter estimation. Besides, the outliers eliminated by screening can reduce the complexity afterwards. (ii) A new distance measure incorporating color factors is proposed to facilitate separating inliers and outliers apart. Taking into consider the color property of most 3D point clouds, the final correctness of outlier removal can be enhanced. (iii) The density-based clustering method is made to be suited to 3D point clouds, being independent of the prior knowledge of the distribution of points. Experimental results on synthetic and real point clouds demonstrate that our approach outperforms the state-of-the-art in both accuracy and computation time.