RoughCough — A new image registration method for radar based vehicle self-localization

摘要

Matching radar gridmap excerpts for vehicle self-localization can be regarded as an image registration task. This paper therefore presents a new, efficient image registration approach, which is suited to radar gridmaps. It does not require sharply structured input images and is applicable to all image pairs that can be aligned by a Euclidean transformation. Few line-shaped parts of the reference image are used for description. The similarity of a test image to these reference lines is calculated in a fast Hough transform based computation scheme. Experiments on radar gridmap excerpts derived from two different test drives demonstrate the low rate of false matches and low registration error of this algorithm. The short computation time shows the suitability of this algorithm for real-time application in a vehicle self-localization setup.