Abstract: Lane marking detection plays crucial role in aided and automatic driving systems as one of the most important parts. This paper mainly studied monocular-vision based lane markings detection. Due to the diversity of lane marking types and the complexity of road condition, efficient and accurate lane marking detection is a challenging problem. In this paper, a novel lane markings detection method is proposed, which introduces deep learning model in the conventional lane marking detection procedure. Our method includes the following steps: first apply prior filter to enhance edges; and then use line segment detector to extract line candidates, followed by a deep convolutional network which can diminish noisy line segments; finally unrelated lines are removed by clustering of vanishing points, and the desired lanes are determined by angle clustering. Empirical results show that the proposed method has high detection accuracy and strong robustness.