A General Lane Detection Algorithm Based on Semantic Segmentation

With the development of artificial intelligent (AI), more and more fields are embracing the AI technology. Intelligent driving is one of the these fields. The environment perception is the fundamental of intelligent driving while lane detection is one of important part of environment perception. Traditional lane detection methods are based on the features of road edge which is only suited on the road with clear car lanes. In this paper, we propose a lane detection method based on semantic segmentation in which includes two stages for lane detection. In the first stage, we use SegNet which is a deep convolutional neural network to recognize the drivable area. Then, we use edge-extracting algorithm to find features of road edge. Based on the extracting edge features, we use cubic curve to fit lane. The experimental results show that our method has a good generalization. It can achieve a good result in both urban roads and rural roads.