Scholarlink

Abstract <p>In this paper, the problem of ensuring automative security control of the subway tunnel is considered. To achieve that, we design and implement a system which automatically detect various subway tunnels relevant events in video sequences. To detect events, the system uses classical computer vision algorithms, such as local image features and connected components analysis. To minimize false event detections and to consider scene changes typical of subway tunnels in a video capture, the system uses integration of individual detections between frames. The event detection quality was measured on the real data of videoclips captured during the subway technical hours, which includes railways inspections by technicians and technical trains passage, and usual subway opening hours. The total precision and recall of event detection in these videos equals to 91.49 and 97.73% respectively. The offline system, designed to optimize the use of computing resources of the device, provides performance at 186 frames per second on the Odroid single-board computer. The proposed method has demonstrated high quality and performance and can be applied on devices with even lower computation power or to ensure subway tunnel security from several video sources simultaneously.</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

Edge Computing Approach for Real-Time Event Detection in Subway Tunnels

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • Daniil Matalov,
  • Sergey Usilin,
  • Vladimir V. Arlazarov

摘要

Abstract

In this paper, the problem of ensuring automative security control of the subway tunnel is considered. To achieve that, we design and implement a system which automatically detect various subway tunnels relevant events in video sequences. To detect events, the system uses classical computer vision algorithms, such as local image features and connected components analysis. To minimize false event detections and to consider scene changes typical of subway tunnels in a video capture, the system uses integration of individual detections between frames. The event detection quality was measured on the real data of videoclips captured during the subway technical hours, which includes railways inspections by technicians and technical trains passage, and usual subway opening hours. The total precision and recall of event detection in these videos equals to 91.49 and 97.73% respectively. The offline system, designed to optimize the use of computing resources of the device, provides performance at 186 frames per second on the Odroid single-board computer. The proposed method has demonstrated high quality and performance and can be applied on devices with even lower computation power or to ensure subway tunnel security from several video sources simultaneously.