Scholarlink

<p>The extraction of vehicle trajectories is essential for understanding traffic flow characteristics and developing effective road traffic management strategies. For this, the aim of this work is to develop an end-to-end model to extract trajectory data from video footage, which includes vehicle detection and classification. As a first step, vehicle detection and classification models were trained using a combination of two existing datasets, IITM-HeTra and FGVD. This approach achieved a mAP50 score of 0.91, which is the best result compared to those obtained using other dataset combinations or individual datasets. Subsequently, vehicle trajectory was extracted using a model trained with YOLOv8, which involved tracking detected vehicles on the screen and obtaining pixel coordinates. The OpenCV library was utilized to convert the obtained pixel coordinates into latitude and longitude data, which were then mapped using Google Earth Pro. The developed model can be applied to any video data to obtain precise vehicle trajectory information, with Google Maps aiding in accurate mapping.</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

Detection, Classification, and Trajectory Extraction of Vehicles under Indian Traffic Conditions

  • 期刊论文

  • 发表日期：

    2026年7月1日

查看全文
  • Ramanand Singh,
  • Abhinav Singh,
  • Ankit Kumar Kushwaha,
  • B. Anil Kumar

摘要

The extraction of vehicle trajectories is essential for understanding traffic flow characteristics and developing effective road traffic management strategies. For this, the aim of this work is to develop an end-to-end model to extract trajectory data from video footage, which includes vehicle detection and classification. As a first step, vehicle detection and classification models were trained using a combination of two existing datasets, IITM-HeTra and FGVD. This approach achieved a mAP50 score of 0.91, which is the best result compared to those obtained using other dataset combinations or individual datasets. Subsequently, vehicle trajectory was extracted using a model trained with YOLOv8, which involved tracking detected vehicles on the screen and obtaining pixel coordinates. The OpenCV library was utilized to convert the obtained pixel coordinates into latitude and longitude data, which were then mapped using Google Earth Pro. The developed model can be applied to any video data to obtain precise vehicle trajectory information, with Google Maps aiding in accurate mapping.