DALNet: A Rail Detection Network Based on Dynamic Anchor Line

Rail detection is one of the key techniques for intelligent train. In this article, we address the problem of detecting rails from an input image. Existing anchor line-based methods typically discretize the continuous space in which all possible rail instances are located into anchor lines at predefined positions. We claim that this commonly used predefined anchor line mechanism is not optimal due to the limitation that it is image-agnostic. Thus we design a novel dynamic anchor line mechanism based on which a rail detection network DALNet is proposed. It utilizes a dynamic anchor line generator (DALG) to dynamically construct an appropriate anchor line for each rail instance based on the position of the rails in the input image. These dynamically generated anchor lines can be regarded as better position references to accurately locate the rails than the predefined anchor lines. In addition, we present a challenging urban rail detection dataset DL-Rail with high-quality annotations and scenario diversity. It is expected to encourage the development of rail detection. We extensively compare DALNet with many fast and competitive lane detection methods. The results show that our DALNet achieves state-of-the-art performance on our DL-Rail rail detection dataset and even on the popular Tusimple and LLAMAS lane detection benchmarks. The code will be released at
https://github.com/Yzichen/mmLaneDet
.