Scholarlink

<p>Multi-target multi-camera tracking (MTMCT) plays a critical role in security surveillance and traffic monitoring applications in AI-driven city environments. However, accurate tracking in dynamic, real-world environments remains difficult. This study focuses on vehicle tracking with MTMCT, employing YOLOv9 as the object detector and enhancing performance by integrating an attention mechanism. The attention mechanism significantly improved tracking accuracy, demonstrating its ability to address challenges posed by diverse and dynamic environments. To evaluate the approach, the system was tested on the CityFlowV2 dataset from the AI City Challenge, organized by NVIDIA, and intersection data from Shin Kong Hospital, provided by the Institute for Information Industry. The system achieved IDF1 scores of 0.8344 and 0.2605 on the CityFlowV2 and Shin Kong Hospital datasets, respectively. Furthermore, the model incorporating GAM achieved the highest IDP score of 0.8929, reflecting the variability in dataset characteristics and highlighting the robustness of the attention mechanism. These results underscore the potential of attention-based enhancements in advancing MTMCT performance and adapting to diverse deployment scenarios.</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

Enhancing multi-target multi-camera vehicle tracking with YOLOv9 and attention mechanisms for smart city traffic monitoring

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Yu-Sheng Tseng,
  • Yu-Fan Su,
  • Daw-Tung Lin

摘要

Multi-target multi-camera tracking (MTMCT) plays a critical role in security surveillance and traffic monitoring applications in AI-driven city environments. However, accurate tracking in dynamic, real-world environments remains difficult. This study focuses on vehicle tracking with MTMCT, employing YOLOv9 as the object detector and enhancing performance by integrating an attention mechanism. The attention mechanism significantly improved tracking accuracy, demonstrating its ability to address challenges posed by diverse and dynamic environments. To evaluate the approach, the system was tested on the CityFlowV2 dataset from the AI City Challenge, organized by NVIDIA, and intersection data from Shin Kong Hospital, provided by the Institute for Information Industry. The system achieved IDF1 scores of 0.8344 and 0.2605 on the CityFlowV2 and Shin Kong Hospital datasets, respectively. Furthermore, the model incorporating GAM achieved the highest IDP score of 0.8929, reflecting the variability in dataset characteristics and highlighting the robustness of the attention mechanism. These results underscore the potential of attention-based enhancements in advancing MTMCT performance and adapting to diverse deployment scenarios.