<p>Due to the special characteristics of the rail transit station scene, common object detection algorithms have false alarms and omissions when detecting targets such as carriers in this scene, and have relatively poor real-time performance. In this paper, we innovatively propose a real-time object detection algorithm combining image depth information and multi-scale attention mechanism, notated as RV-YOLO (Rail Vehicle YOLO). First, RV-YOLO designs the IDIM (image depth information module), which can design the sampling strategy according to the depth information of the image, and in the preprocessing module, the improved backbone module and head module process the image data containing different depth information differently; second, the multi-scale attention module is designed to improve the accuracy of the model in recognizing different angles as well as the occluded targets; then, we propose a new bounding box regression loss function for use in RV-YOLO, which enables the model to gradually focus on more useful information and can compensate for the influence of the shape and scale of the bounding box itself on the regression results as much as possible. Finally, we produced a private dataset (<a href="https://github.com/ChinaZhangPeng/Carrier-Datasets">https://github.com/ChinaZhangPeng/Carrier-Datasets</a>) for the station scenario, which compensates for the lack of the existing public dataset for the rail transit scenario. The experimental results show that RV-YOLO has an AP50 of 97.0% on LVD (Large Vehicle Dataset), 69.8% on MS COCO 2017 dataset, and an FPS of 202 on PASCAL VOC dataset. Compared with object detection models such as YOLOv10 and H-DETR, it provides higher detection accuracy and real-time performance.</p>