A Paired Rail Detection Method Based on Paired Anchors

Rail detection is an important approach to assist train drivers in safe operation. However, existing methods still face challenges in accurately detecting rails in occlusion scenarios. To address this, this paper proposes a paired rail detection approach, which utilizes paired rail priors to extract rail features and iteratively refines rail positions to ultimately predict a pair of rails. The initial rail priors consist of uniformly distributed paired anchors and paired anchors clustered by K-means. Additionally, non-uniform sampling with sparse near the train and dense farther away is employed to improve the regression accuracy of distant rail sections. Finally, to ensure that the classification results directly reflect the quality of regression, the Quality Focal Loss is adopted, which combines the classification task and regression task to provide rail confidence. Our method is evaluated on the DL-Rail dataset. The results demonstrate that our method achieves high detection accuracy, achieving state-of-the-art performance with F1@50 of 91.21 in occluded scenes.