Unsupervised damaged lane restoration for robust lane detection

Abstract

As deep learning technology advances, the development of autonomous driving is accelerating, and the importance of lane detection technology is emerging. However, since various environments are difficult to detect lanes, such as lanes being erased or covered, research is needed to develop a robust lane detection algorithm. This paper proposes an unsupervised lane restoration network that restores damaged lanes for robust lane detection. Because it is not easy to create correct answer data for damaged lanes, the lane restoration network was trained using unsupervised learning. For this, the pre-trained lane detection network was used. To prove the effectiveness of the lane restoration network proposed in this paper, the lane detection performance was measured using the restored image. It was confirmed that the lane detection performance was improved when the proposed lane restoration network was used.