Lane Detection for Autonomous Driving Based on FastSCNN
Boyi Lin · 2024
Lane detection is a key task in automatic driving. In recent years, deep learning methods based on convolutional neural networks have achieved good results in semantic segmentation. In this paper, we analyze the model structure of FastSCNN, a lightweight semantic segmentation network, focus on the performance of FastSCNN in the lane line detection task, and compare it with four other classical models. The experiments prove that FastSCNN is far superior to the other models in terms of real-time performance while ensuring higher accuracy, making it an ideal choice for the lane line detection task. Finally, FastSCNN is successfully deployed to a local PC to simulate in-vehicle real-time lane line detection via a camera with a real-time FPS of 7.37 img/s.