In this paper, we propose a unified end-to-end trainable multi-task networkthat jointly handles lane and road marking detection and recognition that isguided by a vanishing point under adverse weather conditions. We tackle rainyand low illumination conditions, which have not been extensively studied untilnow due to clear challenges. For example, images taken under rainy days aresubject to low illumination, while wet roads cause light reflection and distortthe appearance of lane and road markings. At night, color distortion occursunder limited illumination. As a result, no benchmark dataset exists and only afew developed algorithms work under poor weather conditions. To address thisshortcoming, we build up a lane and road marking benchmark which consists ofabout 20,000 images with 17 lane and road marking classes under four differentscenarios: no rain, rain, heavy rain, and night. We train and evaluate severalversions of the proposed multi-task network and validate the importance of eachtask. The resulting approach, VPGNet, can detect and classify lanes and roadmarkings, and predict a vanishing point with a single forward pass.Experimental results show that our approach achieves high accuracy androbustness under various conditions in real-time (20 fps). The benchmark andthe VPGNet model will be publicly available.
展开▼
