Nighttime lane line detection with fractional differential, Frangi and Hessian

Abstract:

Lane linedetection at night based on machine vision and image processing has always beena research problem in this field, even with the recent deep learning methods, the detection accuracy can only reach to about 50%. To this end, a new algorithm is proposed in this paper. According to the characteristics of the lane line and the speed of the vehicle, multiple images in a video are fused into one detected image; the effective lane line detection area is identified in the region merging by using the characteristics of the image; after the valid detection region is cut into as a new image, the algorithm based on Frag and Hessian matrix is used to smooth and enhance the image; in order to extract the feature points of lane line, a lane line feature point algorithm is proposed based on a new Fractional differential template, then according to the possible position of lane line in the image, the feature points are detected from four directions. After the candidate points are detected, the candidate lane lines are obtained by recursive Hough line transformation. In order to determine the final lane lines,the angle of one lane line should be between 25° and 65°, while the angle of the other lane line should be between 115° and 155°, otherwise, the Hough line transform is continuedby reducing the threshold of the number of lines until two lane lines areobtained. Through testing hundreds of night lane images, the detection accuracy of the new algorithm can reach to 70% compared with deep learning methods and traditional image segmentation algorithms.