Lane detection and curvature estimation based on motion template

We present a lane detection and curvature estimation method by using motion templates. The method is consists of two stages: lane markings are extracted in a top-view image and lane and road curvature are detected based on the lane motion template (LMT). In the first stage, region of interest is selected and transformed into a top-view image, and lane silhouette is simply obtained by convolution with a Gaussian filter. The second stage generates the LMT, which gives motion properties to lane markings by a combination of the lane silhouette and the motion template. To detect lanes, RANSAC algorithm is used to fit a current lane position and historical trajectory with the weight value of the corresponding pixel intensities. Road curvature is estimated by calculating gradient orientations within valid region. In experiments, the proposed method accurately detects the lane markings and road curvature even in a slight curve.)