Scholarlink

<p>Vehicle target detection and tracking is a hot research topic in the field of intelligent transportation. However, due to factors such as complex and variable traffic scenes, occlusion issues, lighting variations, and differences in vehicle scales, accurately detecting and tracking vehicle targets still faces many challenges. This study proposes an improved cascaded matching vehicle detection and tracking algorithm, which is based on the YOLO framework and integrates an optimized feature extraction network with a bidirectional fusion pyramid module, forming a target detection model named YOLO-BFP. To enhance the performance of the model in target tracking, we introduce RIoU module, which provides supplementary information through the height ratio of the detection box to the prediction box, thereby improving the accuracy of tracking association. Furthermore, a lightweight multi-target vehicle tracking framework based on YOLO-BFP was built in this paper. Experimental results on the UA-DETRAC dataset show that the algorithm not only performs outstandingly but also significantly reduces the number of ID switches while maintaining a high processing speed. YOLO-BFP combined with RCTrack achieved HOTA of 59.32% in low-resolution scenarios, outperforming the baseline YOLOv8 + IoU tracker by + 4.46% HOTA<b>,</b> confirming that the lightweight model can maintain high-performance tracking effects while reducing computational burden.</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

Dynamic scale-aware vehicle re-identification via optimized YOLO-BFP and RIoU metric learning

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Xianchen Wang,
  • Can Pei,
  • Jianbiao He,
  • Zhiwei Lu

摘要

Vehicle target detection and tracking is a hot research topic in the field of intelligent transportation. However, due to factors such as complex and variable traffic scenes, occlusion issues, lighting variations, and differences in vehicle scales, accurately detecting and tracking vehicle targets still faces many challenges. This study proposes an improved cascaded matching vehicle detection and tracking algorithm, which is based on the YOLO framework and integrates an optimized feature extraction network with a bidirectional fusion pyramid module, forming a target detection model named YOLO-BFP. To enhance the performance of the model in target tracking, we introduce RIoU module, which provides supplementary information through the height ratio of the detection box to the prediction box, thereby improving the accuracy of tracking association. Furthermore, a lightweight multi-target vehicle tracking framework based on YOLO-BFP was built in this paper. Experimental results on the UA-DETRAC dataset show that the algorithm not only performs outstandingly but also significantly reduces the number of ID switches while maintaining a high processing speed. YOLO-BFP combined with RCTrack achieved HOTA of 59.32% in low-resolution scenarios, outperforming the baseline YOLOv8 + IoU tracker by + 4.46% HOTA, confirming that the lightweight model can maintain high-performance tracking effects while reducing computational burden.