Scholarlink

<p>In sports event scenarios, precise dynamic object detection is critical for real-time monitoring and analysis, yet traditional methods face challenges (small target size, similar appearances, disordered motion, dense occlusions) leading to false detections, low accuracy, and poor robustness. To address these, this paper proposes EAI-YOLO, a real-time dynamic object detection algorithm improved from YOLOv8s: it integrates Efficient Channel Attention (ECA) to enhance small-scale moving target feature capture, replaces traditional Path Aggregation Feature Pyramid Network (PAFPN) with Adaptive Feature Pyramid Network (AFPN) to boost densely occluded object detection, and designs a scale-adaptive Inner-CIoU loss to strengthen generalization for blurry targets and those with abrupt scale changes. Experimental results on a self-made dataset show EAI-YOLO achieves 83.7% mAP@0.5, 45.2% mAP@0.5-0.95, and 0.034s per-frame inference time, balancing real-time performance and detection accuracy. Compared with the YOLOv8s baseline, it increases recall for small objects (e.g., small balls) by 2.5% and detection accuracy for densely occluded objects (e.g., overlapping athletes) by 4.5%, effectively resolving core pain points of dynamic object detection in sports scenarios. This study addresses the limitations of existing methods in complex motion scenarios—insufficient shallow feature extraction and occlusion-induced performance degradation—providing a high-precision technical solution for real-time sports monitoring, athlete motion analysis, and intelligent referee systems, with significant application value for advancing "Smart Venues" construction and the intelligent development of competitive sports.</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

EAI-YOLO:Faster, and more accurate for real-time dynamic object detection

  • 期刊论文

  • 发表日期：

    2025年9月19日

查看全文
  • Jiahao Chen,
  • Kexue Sun,
  • Zhipeng You,
  • Lingqi Xiang

摘要

In sports event scenarios, precise dynamic object detection is critical for real-time monitoring and analysis, yet traditional methods face challenges (small target size, similar appearances, disordered motion, dense occlusions) leading to false detections, low accuracy, and poor robustness. To address these, this paper proposes EAI-YOLO, a real-time dynamic object detection algorithm improved from YOLOv8s: it integrates Efficient Channel Attention (ECA) to enhance small-scale moving target feature capture, replaces traditional Path Aggregation Feature Pyramid Network (PAFPN) with Adaptive Feature Pyramid Network (AFPN) to boost densely occluded object detection, and designs a scale-adaptive Inner-CIoU loss to strengthen generalization for blurry targets and those with abrupt scale changes. Experimental results on a self-made dataset show EAI-YOLO achieves 83.7% mAP@0.5, 45.2% mAP@0.5-0.95, and 0.034s per-frame inference time, balancing real-time performance and detection accuracy. Compared with the YOLOv8s baseline, it increases recall for small objects (e.g., small balls) by 2.5% and detection accuracy for densely occluded objects (e.g., overlapping athletes) by 4.5%, effectively resolving core pain points of dynamic object detection in sports scenarios. This study addresses the limitations of existing methods in complex motion scenarios—insufficient shallow feature extraction and occlusion-induced performance degradation—providing a high-precision technical solution for real-time sports monitoring, athlete motion analysis, and intelligent referee systems, with significant application value for advancing "Smart Venues" construction and the intelligent development of competitive sports.