A lane detection method for freeway aerial videos

This paper proposes a method to detect the lanes from the aerial videos, which is composed of two parts: image pre-processing and the lane detection. The first part, image pre-processing, mainly includes down-sampling, image binarizing and lane edge getting, and then a low resolution binary image will be gotten. The second part consists of the line parameter getting based on Hough Transform, false lane parameters removing, lane parameters correcting based on Kalman algorithm and ROI (region of interest) getting. After taking some characteristics of the freeway into consideration, this paper proposes a binaryzation algorithm which can convert color images into binary images directly. Furthermore, the white blocks in the binary image can be removed after lane edge getting. This paper adopts the Progressive Probabilistic Hough Transform to increase the accuracy and efficiency. What´s more, in this paper, we put forward two principles to remove false line parameters based on the distribution characteristics of the lanes: the dominance of the lane angles and the similarity of the slope difference and the distance between the adjacent lane markings. Finally, in this paper, the Kalman filter is used to correct and predict the parameters. Above all, this method can detect all the lanes correctly from the aerial video, with good accuracy and robustness, even if many vehicles are on the road and pipes and poles are beside the road, and even if the freeway locates at the corner of the image or has a big dip angle.