Dynamic ego-pose estimation for driver assistance in urban environments

Estimating the dynamic ego-pose of the camera system with respect to the road surface is a fundamental task for vision-based driver assistance applications. Existing solutions to the problem are designed to work in well-structured scenarios, i.e., on roads with bright lane markings. This paper presents a stereo-vision approach to the ego-pose estimation problem in urban environments. The proposed algorithm takes the degree of road surface occlusion within the captured traffic scene into account and considers the oscillating characteristics of the camera ego-motion with respect to the road surface. Extensive experiments have shown the feasibility of the algorithm in complex urban environments.