The lane detection is a key step in unmanned vehicles and lane departure warning system. For improving the reliability of lane detection under complicated situation

such as shadows

damaged pavement and vehicle occlusion

the characters of structure lane that the gradient of both side lane marker have opposite directions is utilized

and the lane detection problem is converted to detect the middle line and width of lane based on gradient-pair constraint. Then the parallel perspective model and linear-hyperbola model are used respectively to estimate the vanishing point coordinate

lane width and parameters of middle line by Hough transform. Compared with the other two algorithms under complicated lane situation in extensive experiments

the results demonstrate the effectiveness of the proposed method.