Realtime segmentation of range data using continuous nearest neighbors

In mobile robotics, the segmentation of range data is an important prerequisite to object recognition and environment understanding. This paper presents an algorithm for realtime segmentation of a continuous stream of incoming range data. The method is an extension of the previously developed RBNN algorithm and proceeds in two phases: Firstly, the normal vector of each incoming point is estimated from its neighborhood, which is continuously monitored. Secondly, new points are clustered according to their Euclidean and angular distance to previously clustered points. An outline of the algorithm complexity as well as the parameters that influence the segmentation performance is provided. Three benchmark scenarios in which the algorithm is deployed on a mobile robot with a laser range finder confirm that the method can robustly segment incoming data at high rates.