Point group associations for radar-based vehicle self-localization

Point-shaped objects are a common landmark type for environment-based vehicle self-localization. In this paper, different ways to describe point-shaped landmarks for recognition are investigated: description by absolute position, by features or by constellation in a rigid group with other point-shaped landmarks. Points described by a point group can be associated independently of the prior vehicle pose uncertainty. An efficient algorithm for association of point groups is introduced. Evaluations on real-traffic measurement data from our radar application show the high gain in reliability for associations by point groups and therefore its value for vehicle self-localization tasks.