Multi-Stage and Multi-Scale Lane Detection Under Rainy Weather Conditions

Lane line detection is a crucial aspect of visual technology in autonomous driving. However, in rainy conditions, the noise from rainwater can severely impact lane detection. Therefore, we propose a lane detection network specifically designed for rainy weather scenes. We introduce a feature attention module to learn rainwater features, upon which we build encoding-decoding and global rain removal modules, this facilitates the fusion of spatial details and contextual features. Subsequently, through cross-layer optimization, the network extracts both low-level semantic and high-level semantic information of lane markings, achieving precise lane positioning. Experimental results demonstrate that our method outperforms other algorithmic models under various rainy conditions, effectively addressing the issue of decreased lane detection accuracy in rainy weather.