Hierarchical Clustered Outlier Detection in Laser Scanner Point Clouds

Abstract

Cleaning laser scanner point clouds from erroneous measurements (outliers) is one of the most time consuming tasks that has to bedone before modeling. There are algorithms for outlier detection in different applications that provide automation to some extent butmost of the algorithms either are not suited to be used in arbitrary 3 dimensional data sets or they deal only with single outliers orsmall scale clusters. Nevertheless dense point clouds measured by laser scanners may contain surface discontinuities, noise and diffrentlocal densities due to the object geometry and the distance of the object to the scanner; Consequently the scale of outliers may varyand they may appear as single or clusters. In this paper we have proposed a clustering algorithm that approaches in two steps with theminimum user interaction and input parameters while it can cop with different scale outliers. In the first step the algorithm deals withlarge outliers (those which are very far away from main clusters) and the second step cops with small scale outliers. Since the algorithmis based on clustering and uses both geometry and topology of the points it can detect outlier clusters in addition to single ones. Wehave evaluated the algorithm on a simulated data and have shown the result on some real terrestrial point clouds. The results explainthe potential of the approach to cop with arbitrary point clouds and different scale erroneous measurements.