Robust lane detection based on gradient-pairs constraint

For improving the performance of lane detection in complicated road environment, a novel detection method of structured lane based on gradient-pairs constraint is proposed. After image preprocessing and edge detection, the parametric equation about the mid-line of the road is obtained via Hough Transform with the assumption that a pair of edge pixels on both sides of lane usually has opposite gradient direction; and then, based on the mid-line and edge pixels, the perspective parallel model of the lane is acquired via another Hough Transform; finally, accurate boundary points of the lane can be extracted using the obtained mid-line and the perspective parameters of lane. Through the gradient-pairs constraint as well as twice Hough Transform, the proposed algorithm can overcome the disturbance of shadows, occlusion, artifacts and other cluttered background. By comparing this algorithm with the conventional lane detection algorithm in various road environments, experimental results validate the reliability and effectiveness of the proposed method.