Scholarlink

<p>With the advancement of medical technology, upper limb rehabilitation training methods are gradually shifting towards intelligence and automation. Aiming at the problems of poor adaptability and low robustness of the existing rehabilitation pose estimation, this study proposes an improved Mask R-CNN model. This model optimizes feature extraction by reconstructing residual units and combining attention mechanisms, as well as improving the instance segmentation module to enhance the accuracy and robustness of repair pose estimation, thereby achieving more precise mask prediction. The average accuracy of the improved model was 0.913, the mean accuracy was 0.935, and the frame rate was more than 5.1 ms. The mean square error, root mean square error and mean absolute error (MAE) of the proposed system are 1.3, 1.15 and 0.9, respectively. The proposed method has significantly improved the efficiency, accuracy and robustness of target detection and instance segmentation, and can provide accurate prediction and evaluation for upper limb rehabilitation movement, thus promoting the development of medical rehabilitation evaluation technology to higher quality.</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

Mask R-CNN-based upper limb rehabilitation motion pose estimation method for health monitoring

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月8日

查看全文
  • Xiaoyun Wang

摘要

With the advancement of medical technology, upper limb rehabilitation training methods are gradually shifting towards intelligence and automation. Aiming at the problems of poor adaptability and low robustness of the existing rehabilitation pose estimation, this study proposes an improved Mask R-CNN model. This model optimizes feature extraction by reconstructing residual units and combining attention mechanisms, as well as improving the instance segmentation module to enhance the accuracy and robustness of repair pose estimation, thereby achieving more precise mask prediction. The average accuracy of the improved model was 0.913, the mean accuracy was 0.935, and the frame rate was more than 5.1 ms. The mean square error, root mean square error and mean absolute error (MAE) of the proposed system are 1.3, 1.15 and 0.9, respectively. The proposed method has significantly improved the efficiency, accuracy and robustness of target detection and instance segmentation, and can provide accurate prediction and evaluation for upper limb rehabilitation movement, thus promoting the development of medical rehabilitation evaluation technology to higher quality.