Abstract: A robust lane detection algorithm on unplanar road is proposed. First, a simple inverse perspective mapping method is provided, which is easy to compute and is independent of any camera parameter. Based on the lane model in the normal direction, the line features could be extracted. Then, considering the direction and strength information of extracted line features, the lane line detection could be realized by using improved Hough transform algorithm, which is robust and fast. The detected lane line is the basic lane pattern for further refining the lane points. Finally, the refined lane points are fitted by curves with a weighted least squares method. Experiments demonstrate that the proposed method can detect the lane precisely both on the curve and unplanar road with good robustness.