Scholarlink

<p>The imperative need for accurate traffic data in transportation system operations, management, and planning has driven the search for more efficient data collection methods. This study explores the use of Unmanned Aerial Vehicles (UAVs) combined with deep learning algorithms to enhance vehicle detection, classification, and volume counting. The study employs a DJI Phantom 4 UAV flying at a constant speed of 5&#xa0;m/s at 80&#xa0;m altitude to capture videos, alongside a stationary low oblique angle of 30<sup>0</sup> cameras for analysis. Data from both UAVs and stationary cameras were analyzed using YOLO (you only look once) versions 3, 7, and 8. Leveraging YOLO (you only look once) versions 3, 7, and 8 methodologies, the study employs frame registration, subtraction, thresholding, and morphological operations to address the relative motion between UAV and moving objects, achieving effective vehicle detection, classification, and volume counting. While YOLOv3 demonstrated 100% accuracy for fixed camera data, it struggled with relative motion in UAV data. YOLOv7 overcame this limitation, achieving 95–100% accuracy in classified volume counts for UAV data. Additionally, YOLOv8 enhanced real-time detection precision to 70–100% for fixed camera scenarios. The study highlights the advantages of UAV-based systems, offering broader coverage and adaptability, especially in challenging environments, making them a viable alternative to traditional fixed-camera setups. These findings pave the way for advanced traffic monitoring and data extraction solutions.</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

Estimation of YOLO algorithms for extracting traffic data using UAV and stationary camera

  • 期刊论文

  • 发表日期：

    2025年3月24日

查看全文
  • M. Manjusha,
  • R. Niveditha,
  • V. Sunitha

摘要

The imperative need for accurate traffic data in transportation system operations, management, and planning has driven the search for more efficient data collection methods. This study explores the use of Unmanned Aerial Vehicles (UAVs) combined with deep learning algorithms to enhance vehicle detection, classification, and volume counting. The study employs a DJI Phantom 4 UAV flying at a constant speed of 5 m/s at 80 m altitude to capture videos, alongside a stationary low oblique angle of 300 cameras for analysis. Data from both UAVs and stationary cameras were analyzed using YOLO (you only look once) versions 3, 7, and 8. Leveraging YOLO (you only look once) versions 3, 7, and 8 methodologies, the study employs frame registration, subtraction, thresholding, and morphological operations to address the relative motion between UAV and moving objects, achieving effective vehicle detection, classification, and volume counting. While YOLOv3 demonstrated 100% accuracy for fixed camera data, it struggled with relative motion in UAV data. YOLOv7 overcame this limitation, achieving 95–100% accuracy in classified volume counts for UAV data. Additionally, YOLOv8 enhanced real-time detection precision to 70–100% for fixed camera scenarios. The study highlights the advantages of UAV-based systems, offering broader coverage and adaptability, especially in challenging environments, making them a viable alternative to traditional fixed-camera setups. These findings pave the way for advanced traffic monitoring and data extraction solutions.