Road detection using intrinsic colors in a stereo vision system

Abstract

This thesis describes a vision-based road extraction method for a mobile robot working in an outdoor environment with dynamic lighting changes. We propose a stereo visual sensor system and a long-range road extraction method that is able to accurately detect the drivable road area at distances of up to 50 meters, thus allowing more responsive and efficient path planning. The method is adaptive to different road terrains due to a self-supervised learning process. In addition, it is also shadow-invariant since the learning and classification stages are performed in the intrinsic color space. The color space is designed such that it is representative of intrinsic reflectance of the road surface and independent of illumination source. The advantages of this approach are that it gives more robust results, extends the effective range beyond the stereo range, and, in particular, recognizes shadows on road as drivable road surface instead of non-road areas.