Filtering outliers using statistical analysis on neighbors distances

Laser scanners generally produce point cloud datasets with different point densities. Besides, scanning results are affected by sparse outliers caused by measurement errors. This leads to wrong values when estimate local point cloud features and consequentially upsets point cloud registration. Through statistical analysis on every point and its neighbors, those outliers can be found out. We compute the distances from points to neighbors, and get the distribution of the mean distance. Assuming that the resulted distribution is Gaussian with a mean and a standard deviation, we find out outliers and delete them from the dataset, because their mean distances are outside the district decided by the distances expectation and standard deviation.