Filtering algorithm for LiDAR Outliers based on histogram and KD tree

LiDAR´s outliers include points distinctly higher or lower than their surroundings and isolated points, which are normally caused by birds, low flying aircrafts, multi-path errors and system errors. It´s necessary to remove LiDAR´s outliers before classifying LiDAR ground points. In this study, laser points´ elevations are transformed into a histogram from 0 to 255 elevation scales. Then, the histogram is split by some thresholds with a multilevel segmentation algorithm. A small amount of higher or lower laser points, as they are located at the starting or ending part of the histogram, are filtered into Low Point (noise) class refer to point proportion threshold. In the next step, the algorithm creates an unclassified laser points KD tree and searches the number of points around each laser point in given querying radius. If the number is less than a given point number threshold after increasing search radius length several times, the point is treated as isolated point, i.e. Low Point (noise) class. In experiments, it is shown that this filtering algorithm is reliable in filtering LiDAR´s outliers.