A parallel memory efficient outlier detection algorithm for large unstructured point clouds

The removal of erroneous points is an important pre-processing step in the analysis of point clouds, to ensure a high quality evaluation of the measured objects. However, modern optical 3D scanning technologies generate point clouds that can contain hundreds of millions of points. Existing algorithms for the removal of erroneous points can face difficulties, due to the amount of memory that is required to process these point clouds. We present a new method that is based on the well known Local Outlier Factor algorithm. It adapts the calculation of the factor in minor ways to reduce its runtime. More importantly, we employ a new processing strategy that significantly reduces the overall memory consumption of the algorithm. This enables the detection and removal of outliers even in very large point clouds. In order to show the effectiveness of our new method, we evaluate the processing on multiple, differently sized point sets and demonstrate the configurable memory consumption of our new technique.