Deep Learning Based Lane Line Detection and Segmentation Using Slice Image Feature

Abstract

Nowadays, an effective driving assist system is expected to perform fast to observe and taking immediate decision. In particular for the detection of lane lines the system should able to perform faster and accurately locate the position of the lane lines. The majority of the existing work in this task relies on the frame-based processing in which the whole image is used as a feature. In addition for the case of high-resolution images the computational time is very significant and not feasible for practical applications in particular for embedded system. To overcome this problems, in this work a novel lane line detection system is proposed. The proposed approach utilizes the slice of a frame image as a feature and applies deep learning to detect and segment the lane line. Experimental results show that the proposed system can efficiently handle lane line detection by 60 times faster than the former schemes with superior accuracy.