Scholarlink

<p>Leveraging advanced image processing techniques, traffic management surveillance systems efficiently analyze vehicle trajectories and document traffic violations. Enhanced by big data analytics and Artificial Intelligence (AI), the law enforcement agencies achieve heightened operational efficiency in addressing such incidents. The integration of Intelligent Transportation Systems (ITS) with data-driven decision support platforms substantially elevates traffic safety oversight. This technological synergy safeguards urban mobility and promotes the sustainable advancement of smart cities. To improve street-level surveillance efficacy specifically for non-motorized vehicle violations, this study introduces a two-dimensional distributed visual analysis architecture based on the You Only Look Once version 7 (YOLOv7). The module dynamically adapts through rapid learning based on target dimensions (height, width) and velocity characteristics to enhance detection performance. The distributed structure is optimized by incorporating the Deep Simple Online and Real-time Tracking (Deep SORT), which eliminates constraints related to external feature map dimensions through the integration of adjustment factors. Further refinements to the image processing methodology augment the detection capability for small targets. Multi-scale feature fusion is enhanced by introducing sampling layers into the external feature integration process, complemented by skip connections that facilitate the effective fusion of high-level and low-level semantic information. The experimental analysis demonstrates that the proposed method achieves a recognition accuracy of 95.2%. Both ablation studies and visual assessments confirm its superior performance compared to existing approaches. These results validate that the research achievements enhance model recognition accuracy, provide effective early warnings for non-motorized vehicle violations, and contribute to reducing roadway safety incidents.</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

Non-motor vehicle violation real-time prediction system based on YOLOv7-deep SORT street view video

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Yubian Wang,
  • Zhixu Luo,
  • Aleksandr Aleksandrovich Erofeev

摘要

Leveraging advanced image processing techniques, traffic management surveillance systems efficiently analyze vehicle trajectories and document traffic violations. Enhanced by big data analytics and Artificial Intelligence (AI), the law enforcement agencies achieve heightened operational efficiency in addressing such incidents. The integration of Intelligent Transportation Systems (ITS) with data-driven decision support platforms substantially elevates traffic safety oversight. This technological synergy safeguards urban mobility and promotes the sustainable advancement of smart cities. To improve street-level surveillance efficacy specifically for non-motorized vehicle violations, this study introduces a two-dimensional distributed visual analysis architecture based on the You Only Look Once version 7 (YOLOv7). The module dynamically adapts through rapid learning based on target dimensions (height, width) and velocity characteristics to enhance detection performance. The distributed structure is optimized by incorporating the Deep Simple Online and Real-time Tracking (Deep SORT), which eliminates constraints related to external feature map dimensions through the integration of adjustment factors. Further refinements to the image processing methodology augment the detection capability for small targets. Multi-scale feature fusion is enhanced by introducing sampling layers into the external feature integration process, complemented by skip connections that facilitate the effective fusion of high-level and low-level semantic information. The experimental analysis demonstrates that the proposed method achieves a recognition accuracy of 95.2%. Both ablation studies and visual assessments confirm its superior performance compared to existing approaches. These results validate that the research achievements enhance model recognition accuracy, provide effective early warnings for non-motorized vehicle violations, and contribute to reducing roadway safety incidents.