<p>As the artificial intelligence and sensing technology rapid develop, accurate recognition of complex sports such as volleyball is greatly significant for improving training efficiency and event analysis. In response to the accuracy and efficiency issues of traditional camera technology in the special environment of volleyball sports, this study first uses millimeter wave radar to collect point cloud spatial data of athletes, and then extracts human spatial features from the data through convolutional neural networks to raise the resolution and accuracy of the data. Next, a long short-term memory network is used to process time-series data and capture the spatiotemporal changes in volleyball technique movements. The results indicate that the millimeter wave radar used in the study has a delay of only 0.55 × 10<sup>−6</sup>s and has good accuracy. When processing a single frame of data, the convolutional neural network has significant errors in extracting joints 7 and 10 in the X-axis direction, with errors of 11.0 and 10.6, respectively. The remaining joint errors are all below 10, indicating that this model can effectively extract image features and estimate the joint positions of human movements. The proposed model performs outstandingly in outdoor volleyball environments, although its response time is slightly slower than that of LiDAR, its accuracy and memory requirements are significantly better than other systems. Its accuracy in outdoor volleyball environment reaches 95.6%, with a response time of only 2.9&#xa0;ms, demonstrating higher accuracy and lower computational delay than traditional methods. This approach offers a novel method for the precise recognition of volleyball technical movements, while also playing a pivotal role in enhancing the analysis of movements within the volleyball sports environment.</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

Volleyball technical action recognition based on CNN-LSTM

  • Zhigang Zhang,
  • Yong Tian,
  • Jinchong Qi

摘要

As the artificial intelligence and sensing technology rapid develop, accurate recognition of complex sports such as volleyball is greatly significant for improving training efficiency and event analysis. In response to the accuracy and efficiency issues of traditional camera technology in the special environment of volleyball sports, this study first uses millimeter wave radar to collect point cloud spatial data of athletes, and then extracts human spatial features from the data through convolutional neural networks to raise the resolution and accuracy of the data. Next, a long short-term memory network is used to process time-series data and capture the spatiotemporal changes in volleyball technique movements. The results indicate that the millimeter wave radar used in the study has a delay of only 0.55 × 10−6s and has good accuracy. When processing a single frame of data, the convolutional neural network has significant errors in extracting joints 7 and 10 in the X-axis direction, with errors of 11.0 and 10.6, respectively. The remaining joint errors are all below 10, indicating that this model can effectively extract image features and estimate the joint positions of human movements. The proposed model performs outstandingly in outdoor volleyball environments, although its response time is slightly slower than that of LiDAR, its accuracy and memory requirements are significantly better than other systems. Its accuracy in outdoor volleyball environment reaches 95.6%, with a response time of only 2.9 ms, demonstrating higher accuracy and lower computational delay than traditional methods. This approach offers a novel method for the precise recognition of volleyball technical movements, while also playing a pivotal role in enhancing the analysis of movements within the volleyball sports environment.