Scholarlink

<p>Table tennis has the characteristics of small size and high speed, and existing object detection algorithms have low recognition accuracy. To address the problem of detecting and predicting the trajectory of small objects in high-speed motion, this paper presents an intelligent prediction model for small object motion trajectory based on ShuffleNet and YOLOv5s. The model reduces parameter size by using grouped convolution and channel shuffle. It combines the dual-stage association matching of ByteTrack and applies an unscented transform to reduce the interference of motion blur in small objects. It also uses a Gated Recurrent Unit to capture temporal dependence, so as to achieve adaptive tracking and prediction of table tennis trajectory. In test experiments, the average classification accuracy of this model is 98.05%, the detection speed is 522.5 fps, and the landing point detection accuracy is 98.21%, which is significantly higher than those of the compared models. These results show that the model has more efficient lightweight computing ability, meets the demand of real-time detection and trajectory tracking of table tennis in high-speed motion, and provides reliable algorithm support for table tennis training and match data analysis.</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

Table tennis detection and trajectory prediction based on shuffle-YOLOv5s algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月29日

查看全文
  • Yuxue Wang,
  • Yingying Yang

摘要

Table tennis has the characteristics of small size and high speed, and existing object detection algorithms have low recognition accuracy. To address the problem of detecting and predicting the trajectory of small objects in high-speed motion, this paper presents an intelligent prediction model for small object motion trajectory based on ShuffleNet and YOLOv5s. The model reduces parameter size by using grouped convolution and channel shuffle. It combines the dual-stage association matching of ByteTrack and applies an unscented transform to reduce the interference of motion blur in small objects. It also uses a Gated Recurrent Unit to capture temporal dependence, so as to achieve adaptive tracking and prediction of table tennis trajectory. In test experiments, the average classification accuracy of this model is 98.05%, the detection speed is 522.5 fps, and the landing point detection accuracy is 98.21%, which is significantly higher than those of the compared models. These results show that the model has more efficient lightweight computing ability, meets the demand of real-time detection and trajectory tracking of table tennis in high-speed motion, and provides reliable algorithm support for table tennis training and match data analysis.