This paper proposes some methods of increasing the robustness and accuracy of lane detection in bad weather, such as light rain or after a rain. First, a reflection model of road surface water is presented. The model can efficiently fit the real reflection phenomenon on a rainy road with strong specular reflection of sky light on the surface water or road, and it can efficiently reduce specular reflection and enhance lane markings in the same situation. Second, the paper proposes some methods of extracting lane markings based the symmetry feature. They include the methods of detecting vanishing point and fitting the shape model of the parallel lanes using the symmetry features. Finally, by combining some prior knowledge, such as restriction of slope of the parallel lane lines and symmetry features, the methods can effectively increase the stability, robustness and accuracy of lane detecting in different complex environments.