Scholarlink

<p>The railroad industry has witnessed a growing need for accurate train position data to enhance operational efficiency and safety, leading to extensive research in this area. To improve train positioning, we propose a vision-based system using a YOLO v3 object detection neural network to identify overhead catenary poles, a railway facility, as landmarks on the side of railroad tracks. The system estimates the position of a train by measuring the distance to the closest catenary pole. To ensure image data stability, the proposed positioning system utilizes sensor fusion by combining the vision-based results with onboard odometer measurements using a Kalman filter. We used a GPS to evaluate the proposed image-based positioning system and verified its stability, even under conditions of train slipping or sliding. Comparing the performance to GPS ground truth, the proposed vision-based method outperformed the existing odometer-based approach. Over a 770&#xa0;m test route, the vision-based system achieved a maximum error of 1.57&#xa0;m and RMSE of 0.71&#xa0;m. This study highlights the potential of vision-based train positioning technology to enhance railroad operational efficiency and safety. The use of stable overhead catenary poles as landmarks, combined with sensor fusion techniques, demonstrates an alternative approach to improving train localization.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of Vision-Based Train Positioning System Using Object Detection and Kalman Filter

  • 期刊论文

  • 发表日期：

    2025年1月13日

查看全文
  • Hyun Jung Kim,
  • Yeun Sub Byun,
  • Rag Gyo Jeong

摘要

The railroad industry has witnessed a growing need for accurate train position data to enhance operational efficiency and safety, leading to extensive research in this area. To improve train positioning, we propose a vision-based system using a YOLO v3 object detection neural network to identify overhead catenary poles, a railway facility, as landmarks on the side of railroad tracks. The system estimates the position of a train by measuring the distance to the closest catenary pole. To ensure image data stability, the proposed positioning system utilizes sensor fusion by combining the vision-based results with onboard odometer measurements using a Kalman filter. We used a GPS to evaluate the proposed image-based positioning system and verified its stability, even under conditions of train slipping or sliding. Comparing the performance to GPS ground truth, the proposed vision-based method outperformed the existing odometer-based approach. Over a 770 m test route, the vision-based system achieved a maximum error of 1.57 m and RMSE of 0.71 m. This study highlights the potential of vision-based train positioning technology to enhance railroad operational efficiency and safety. The use of stable overhead catenary poles as landmarks, combined with sensor fusion techniques, demonstrates an alternative approach to improving train localization.