Lane Marking Detection Techniques for Autonomous Driving

Abstract

In this paper we address challenges facing lane marking detection and tracking. Lane marking detection along with vehicle positioning between lane boundaries are fundamental tasks to achieve safe and reliable autonomous driving systems. Despite the development of perception senors and clarity of the lane markings on roadways, the lane detection remains a challenge for researchers due to environmental factors that impact performance of lane's recognition algorithms. In this paper we compare three different lane detection strategies based on rule and learning-based approaches using perception sensors. In contrast, we perform rule-based lane detection using camera and sensor fusion combining camera images and LiDAR's point clouds. Moreover, we use the prominent lane detection learning-based approach LaneNet to detect lanes from images. However, when using the LaneNet, we investigate the network's performance while excluding the perspective transformation network (H-Net). Our results show that learning-based lane detection methodologies outperforms rule-based methods and can accurately predict lanes when the vehicle is cruising on roads with steep curvatures.