Vision Based Road Crossing Scene Recognition for Robot Localization

Abstract

An approach of road crossing scene recognition based on scale invariant feature transform (SIFT) and color features is proposed in this paper. Firstly, the SIFT features are extracted and the color histogram in HSI space is calculated. Secondly, the K-D trees algorithm is used to match SIFT features of images in road crossing images database, and Bhattacharyya distance match result is calculated by color histogram. Finally, the SIFT features match result and Bhattacharyya distance match result are combined together to confirm the suitable image in database. The image pre-classified idea is also adopted to accelerate the SIFT features matching. The experiment results demonstrate that the algorithm is robust to the various illumination, dynamic disturbance and self-circumrotating, and can be used to the robot location.