Scholarlink

<p>In this paper, a real-time tracking model is proposed to address the low accuracy and poor identity stability in multi-target tracking caused by high-speed movement, frequent occlusion, and similar appearances between players and volleyball targets in volleyball match videos. The model deeply fuses the improved YOLOv11 object detector with the enhanced OC-SORT multi-object tracker. In the detection stage, through scene-adaptive feature fusion and lightweight design, the model achieves a mean average precision of 96.5% on the self-built volleyball dataset and a processing speed of 142 FPS. In the tracking stage, the nonlinear motion and complex occlusion problems are effectively addressed by introducing a motion model-adaptive mechanism and an occlusion inference sub-network. Comprehensive experiments on public data sets and self-collected videos demonstrate the clear quantitative superiority of ST-Net; it achieves 87.2% on the key indicator MOTA and successfully strikes an optimal balance between accuracy and speed with an overall tracking speed of 38 FPS, while significantly enhancing identity stability by reducing identity switches to an average of only 3.1 times per game. These results explicitly verify that the SpikeTrack-Net algorithm offers prominent advantages in high precision, exceptional stability, and strong real-time performance in volleyball match scenes, providing a highly reliable technical scheme for the automatic analysis of sports techniques and tactics.</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

Research on multi-target real-time tracking algorithm of volleyball match based on YOLOv11 and OC-SORT

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月20日

查看全文
  • Dan Lv,
  • Bo Bai

摘要

In this paper, a real-time tracking model is proposed to address the low accuracy and poor identity stability in multi-target tracking caused by high-speed movement, frequent occlusion, and similar appearances between players and volleyball targets in volleyball match videos. The model deeply fuses the improved YOLOv11 object detector with the enhanced OC-SORT multi-object tracker. In the detection stage, through scene-adaptive feature fusion and lightweight design, the model achieves a mean average precision of 96.5% on the self-built volleyball dataset and a processing speed of 142 FPS. In the tracking stage, the nonlinear motion and complex occlusion problems are effectively addressed by introducing a motion model-adaptive mechanism and an occlusion inference sub-network. Comprehensive experiments on public data sets and self-collected videos demonstrate the clear quantitative superiority of ST-Net; it achieves 87.2% on the key indicator MOTA and successfully strikes an optimal balance between accuracy and speed with an overall tracking speed of 38 FPS, while significantly enhancing identity stability by reducing identity switches to an average of only 3.1 times per game. These results explicitly verify that the SpikeTrack-Net algorithm offers prominent advantages in high precision, exceptional stability, and strong real-time performance in volleyball match scenes, providing a highly reliable technical scheme for the automatic analysis of sports techniques and tactics.