Scholarlink

<p>Dual modal data of rail image and axle box acceleration are crucial for railway maintenance and repair. Due to the different collection equipment and sampling intervals of the raw data, there are alignment errors. Data alignment is the foundation and prerequisite for multimodal data fusion detection of rails. To address the misalignment between rail image and acceleration data, this paper proposes a method for aligning rail image and axle box acceleration based on variable-scale pattern matching. This method uses rail joint feature points as the basis for alignment, transforming the dual modal data alignment problem into a pattern matching problem between image and acceleration feature point sequences. By constructing variable-scale patterns, the feature points are matched separately in different patterns, and the results are relatively calibrated based on uncertainty. This effectively reduces error accumulation and improves data alignment accuracy. Experimental results show the generalizability and robustness of the proposed method. The mean absolute error at rail joints is less than 0.02&#xa0;m, while at the point rails it remains within 0.5&#xa0;m. Our source code is publicly available at: <a href="https://github.com/ccyppl/VSAA">https://github.com/ccyppl/VSAA</a>.</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

A dual modal alignment method of rail image and axle box acceleration based on variable-scale pattern matching

  • 期刊论文

  • 发表日期：

    2026年2月5日

查看全文
  • Changlun Zhang,
  • Guocui Zhang,
  • Yu Cheng,
  • Qiang He,
  • Hengyou Wang,
  • Jinzhao Liu

摘要

Dual modal data of rail image and axle box acceleration are crucial for railway maintenance and repair. Due to the different collection equipment and sampling intervals of the raw data, there are alignment errors. Data alignment is the foundation and prerequisite for multimodal data fusion detection of rails. To address the misalignment between rail image and acceleration data, this paper proposes a method for aligning rail image and axle box acceleration based on variable-scale pattern matching. This method uses rail joint feature points as the basis for alignment, transforming the dual modal data alignment problem into a pattern matching problem between image and acceleration feature point sequences. By constructing variable-scale patterns, the feature points are matched separately in different patterns, and the results are relatively calibrated based on uncertainty. This effectively reduces error accumulation and improves data alignment accuracy. Experimental results show the generalizability and robustness of the proposed method. The mean absolute error at rail joints is less than 0.02 m, while at the point rails it remains within 0.5 m. Our source code is publicly available at: https://github.com/ccyppl/VSAA.