Liu, Yongqiang (Beijing jiaotong university), Jiang, Wei (Beijing Jiaotong University), Hu, Xiao (International Digital Economy Academy), Wang, Jian (Beijing Jiaotong University), Cai, Bai-gen (Beijing Jiaotong University)

A Seamless Train Positioning Method Based on Visual Place Recognition

Scheduled for presentation during the Regular Session "Sensing, Vision, and Perception V" (FrAT5), Friday, September 27, 2024, 12:10−12:30, Salon 13

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Accurate and reliable train positioning stands as a cornerstone in railway train control system applications. In this paper, a seamless train positioning method based on visual position identification (VPR) is proposed. Our method uses on-board vision sensors to continuously capture images of the railway environment. Then, these images are processed by using the NetVLAD model to extract visual features. Subsequently, Euclidean distance is employed to conduct similarity retrieval on these features against pre-constructed railway image maps, facilitating precise positioning along the train line. This method has several obvious advantages. First, independence from external infrastructure: Unlike traditional methods reliant on trackside infrastructure such as physical transponders, our approach achieves precise positioning akin to a transponder without external dependencies. Second, decoupled position processes: Each image retrieval process operates independently, devoid of strong coupling relationships between positioning at adjacent frames. Third, robust adaptability to diverse environmental conditions: Experiments show that train location can be achieved in both open environment and tunnel scenes, and the coverage rate of effective train location reaches 98.2%, which proves the potential of this method in train location.