Lane detection is a crucial component of many advanced driver-assistance systems. Lane position can help both develop many driving assistance systems such as lane keeping system (LKS), lane departure warning system (LDWS) and achieve vehicle local location and behavior prediction. In this paper, an illumination invariant lane detection method was proposed. The proposed method takes full use of time-space context for robust lane detection. First, the vanishing point was detected adaptively in the strip image by using standard Hough transform. Thus, the region of interest (ROI) can be obtained adaptively for computation reduction. Then, considering the time-context information of the image sequence, a procedure for generating the time slice image which is used to acquire the lane marks key points was designed. In addition, Tophat processing was utilized to enhance contrast degree of the time slice image for better segmentation between the lane marks and road under the noise and low illumination. Finally the method for predicting the current lane marks position was given.Several driving scenarios had been tested and the experiment results show that the proposed lane detection method works well with acceptable performance and has enough tolerance to low illumination, variant illumination, broken and blurredlanes.
展开▼
