Traffic road line detection based on the vanishing point and contour information

This paper describes an algorithm to detect the road lane boundary for a driver who changes lane as `access road´, `passing lane´, `approach ramp´, etc. The proposed algorithm starts with binarization and labeling for the image. The binary image is generated by adaptive threshold. The binary image is labeled recursively to blobs. From each blob, the algorithm analyses principal axes calculated by Eigen value and vector. Those are based on the covariance matrix of pixel distribution for the blobs. Then it shows intersection points by the combination of each long principal axis. Among intersection points, the highest density cluster is selected to a dominant vanishing point as the parallel road lines. In experiments of road line detection from the urban road image, the proposed method shows the effectiveness of lane boundary detection in the rate of more than 94.3%.