Scholarlink

In real-world floorball event scenarios, players exhibit large movement ranges, diverse postural changes, and intense physical confrontations, often leading to occlusions between players. Traditional object detection models are prone to false detections and missed detections, especially when occlusion occurs, which can significantly impact detection performance. To address this issue and improve target recognition accuracy, this study proposes a new object detection model, CCD-YOLO11n, based on the YOLO11n framework. The proposed model introduces ConvNeXtV2 as the backbone network, leveraging its larger receptive field and stronger feature extraction capability. Meanwhile, the Coordinate Attention (CA) mechanism is integrated into the C3k2 structure of YOLO11 to form a novel C3k2_CA module, which further enhances the model's ability to focus on critical information, thereby improving recognition accuracy in complex and occluded scenarios. Moreover, DySample is introduced as the upsampling module in CCD-YOLO11n. Its efficient dynamic sampling mechanism and lightweight design enable high-quality dynamic conversion from low-resolution to high-resolution feature maps, further optimizing feature fusion and enhancing overall performance. To better detect player in floorball, we have carefully constructed the Floorball-PD dataset, which contains 4,268 real floorball game scene images. Experimental results show that our model improve YOLO11n's mAP@0.5 by 2.1% and mAP@0.5:0.95 by 1.9%.

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

CCD-YOLO11n: A Novel Approach for Player Detection in Floorball

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Bing Xu,
  • Junwei Mu,
  • Yongmei Liu,
  • Ming Ma

摘要

In real-world floorball event scenarios, players exhibit large movement ranges, diverse postural changes, and intense physical confrontations, often leading to occlusions between players. Traditional object detection models are prone to false detections and missed detections, especially when occlusion occurs, which can significantly impact detection performance. To address this issue and improve target recognition accuracy, this study proposes a new object detection model, CCD-YOLO11n, based on the YOLO11n framework. The proposed model introduces ConvNeXtV2 as the backbone network, leveraging its larger receptive field and stronger feature extraction capability. Meanwhile, the Coordinate Attention (CA) mechanism is integrated into the C3k2 structure of YOLO11 to form a novel C3k2_CA module, which further enhances the model's ability to focus on critical information, thereby improving recognition accuracy in complex and occluded scenarios. Moreover, DySample is introduced as the upsampling module in CCD-YOLO11n. Its efficient dynamic sampling mechanism and lightweight design enable high-quality dynamic conversion from low-resolution to high-resolution feature maps, further optimizing feature fusion and enhancing overall performance. To better detect player in floorball, we have carefully constructed the Floorball-PD dataset, which contains 4,268 real floorball game scene images. Experimental results show that our model improve YOLO11n's mAP@0.5 by 2.1% and mAP@0.5:0.95 by 1.9%.