A robust lane boundaries detection algorithm based on gradient distribution features

The paper presents a lane boundaries detection algorithm based on gradient distribution features. Firstly, a combination of EDF and Hough transform is used to obtain the linear models of lane boundaries. Secondly, without any prior knowledge, the width of road, the middle point of road and the other parameters are computed based on the linear models. Finally, bi-directional sliding window technique is applied to detect real lane markings. Experimental results indicate that the proposed method can enhance the adaptability to deal with the random and dynamic environment of road scenes, such as curved lane markings, sparse shadows, object occlusions and bad conditions of road painting.