Scholarlink

<p>The detection of obstacles at railway level crossings is crucial for ensuring the safety of passengers and cargo, as well as for maintaining a smooth flow of road and rail traffic. In this context, we present a deep learning-based video analytics algorithm tailored for deployment on smart cameras, able to autonomously detects the presence of individuals and vehicles at railway level crossings. To mitigate false alarms triggered by objects when the level crossing is open (thus the passage of objects is allowed), the proposed algorithm also incorporates a deep neural network to automatically determine the barrier status (open or closed). The proposed system has been evaluated on a data set of 52 videos of railway crossings we collected from Youtube, exhibiting impressive 0.98 precision and 0.88 recall. In addition, the proposed system is designed to operate directly on smart cameras and embedded devices, eliminating the need for server infrastructure or cloud connectivity.</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

Improving safety through advanced obstacle detection at railway level crossings

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Vincenzo Carletti,
  • Antonio Greco,
  • Alessia Saggese,
  • Bruno Vento

摘要

The detection of obstacles at railway level crossings is crucial for ensuring the safety of passengers and cargo, as well as for maintaining a smooth flow of road and rail traffic. In this context, we present a deep learning-based video analytics algorithm tailored for deployment on smart cameras, able to autonomously detects the presence of individuals and vehicles at railway level crossings. To mitigate false alarms triggered by objects when the level crossing is open (thus the passage of objects is allowed), the proposed algorithm also incorporates a deep neural network to automatically determine the barrier status (open or closed). The proposed system has been evaluated on a data set of 52 videos of railway crossings we collected from Youtube, exhibiting impressive 0.98 precision and 0.88 recall. In addition, the proposed system is designed to operate directly on smart cameras and embedded devices, eliminating the need for server infrastructure or cloud connectivity.