Abstract
Deep-Learning-Based algorithms successfully improve the detection accuracy of traffic lanes. Most of them employed stacked convolutional neural networks( CNNs) to extract semantics, but they have trouble gathering global visual information. What’s more, the stacked CNNs perform multiple down-sampling, which reduce the resolution of feature map, causing the tiny features easy to be ignored. To address these problems, LaneTD is proposed: a novel lane feature aggregator that extracts and fuses local and global features. This method adopts dilated convolution to extract local features of various scales, and enhances global representation base on transformer encoder. The local and global information are then weighted fused using bidirectional feature pyramid. In addition, we apply unsupervised image style transfer to improve the performance in nighttime and dark scenarios. The method is quantitatively evaluated on public challenging CULane dataset, the result shows that the proposed method significantly improves the accuracy and computational efficiency of lane detection in different scenarios. Furthermore, an ablation study is carried out, as well as a discussion of efficiency trade-off choices that can be applied in practice.
Funding Information
  • Research and Development Program of Key Science and Technology Fields in Guangdong Province (2019B090912001)
  • Guangzhou Key Field Research and Development Program (202206030005)