Outlier Detection in Laser Scanner Point Clouds

Outlier detection in laser scanner point clouds is an essential process before the modelling step. However, the number of points in the generated point cloud is in the order of million points, so (semi) automatic approaches are necessary. Having introduced the sources of outliers in typical laser scanner point clouds, an outlier detection algorithm using a density based algorithm is addressed. The algorithm is chosen due to its unconstrained behaviour to the preliminary knowledge of the scanned scene and the independency to the varying density of the points. The algorithm efficiency is assessed by a test on an aerial laser scanner point cloud. The assessment is done with respect to a DSM obtained by photogrammetric methods. The type I and type II errors are estimated and the results are reported. Some examples in terrestrial laser scanner point clouds are also presented and the behaviour of the algorithm on the data sets are shown and discussed. Although the algorithm dose not detect all outlier clusters, the detection of single outliers and even cluster outliers with lower density than a predefined value seems satisfactory, specially the results in the boundary of occlusions. Detection of such outliers is important because if they remain in the data set, they may cause modelling errors in the next modelling step.