Scholarlink

Multiple object tracking (MOT) is a significant component in vision-based navigation and surveillance systems. This study presents a practical integration of YOLOv4 for object detection, Kalman filters for motion prediction, and the Hungarian algorithm for object association to build a robust MOT framework. The system effectively handles occlusions and object re-identification by maintaining consistent object IDs across video frames. Evaluation on COCO-derived video sequences demonstrates improved ID stability and accuracy compared to YOLOv4 alone, with a trade-off in processing speed. While the approach lacks architectural novelty, it highlights the efficiency of combining well-established algorithms for real-time applications. Future work includes optimizing speed and benchmarking with more advanced trackers such as DeepSORT and FairMOT.

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

Improvisation of Yolov4 Algorithm for Multiple Object Tracking Using Kalman Filtering Hungarian Method

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • H. Faizal Ahamed,
  • M. Brindha,
  • G. Sri Sowmiya Narayanan

摘要

Multiple object tracking (MOT) is a significant component in vision-based navigation and surveillance systems. This study presents a practical integration of YOLOv4 for object detection, Kalman filters for motion prediction, and the Hungarian algorithm for object association to build a robust MOT framework. The system effectively handles occlusions and object re-identification by maintaining consistent object IDs across video frames. Evaluation on COCO-derived video sequences demonstrates improved ID stability and accuracy compared to YOLOv4 alone, with a trade-off in processing speed. While the approach lacks architectural novelty, it highlights the efficiency of combining well-established algorithms for real-time applications. Future work includes optimizing speed and benchmarking with more advanced trackers such as DeepSORT and FairMOT.