An Effective Module CA-HDC for Lane Detection in Complicated Environment

As an important part of intelligent vehicle, some lane detection algorithms based on deep learning have achieved higher accuracy compared with traditional computer vision. Applied in some scenes with good traffic environment, these algorithms perform excellent. However, lane detection in complicated environment is still having a lot of room to improve, especially in the night, crowded and other complicated scenes. Low accuracy of lane detection may cause serious traffic accidents. This paper proposes a lane detection algorithm using hybrid dilated convolution and channel attention mechanism (CA-HDC) to improving the performance of deep learning network. This algorithm includes a feature extraction layer to extracting lane feature information and three detection branches to realize location and segmentation. Experiments on the TuSimple and CULane dataset show that the proposed network outperforms the baseline.