Fast and Accurate Road Region Detection Techniques

Abstract

By using an onboard camera, it is possible to detect the free road surface ahead of the vehicle. Road detection is of high relevance for road departure warning, autonomous driving and supporting driver-assistance systems such as vehicle and pedestrian detection. Significant efforts have been made in order to solve this task using vision based techniques. One of the major challenges of these techniques is dealing with lighting variations, especially extreme shadows and highlights. We present an innovative method to obtain a road segmentation algorithm robust to extreme shadow and highlight conditions. The novelty of our approach is that we combine the shadow-invariant feature space with horizon estimation, tone mapping and the probability density function (PDF)-based classification to achieve fast and accurate road region segmentation. The state-of-the-art method uses likelihood-based classifier to select threshold by assuming the road region, so this method is time-consuming and produces inaccurate results. In our approach we use PDF to directly estimate threshold of road pixels. Horizon estimation is significantly important in reducing computational time and improving accuracy. Tone mapping could avoid the influence of highlights and penumbra caused by cameras of lower dynamic range. By this, we achieve fast and accurate segmentation of the input image into road and non-road regions. Moreover, the proposed algorithm works in still images and does not depend on either road shape or temporal restrictions.