Scholarlink

<p>Rehabilitation following surgery or trauma is crucial for patient recovery. However, many patients attempt to rehabilitate at home by performing exercises without professional supervision, potentially hindering their progress or causing complications. To address this, deep learning systems have been extensively explored. Traditional motion detection systems such as Vicon or Kinect, while effective, are costly or impractical for at-home use. Therefore, leveraging the ubiquity of mobile devices, we propose an affordable and robust home-based rehabilitation system that uses a simple RGB camera for recording physical exercises. We used Mediapipe, an open-source framework, to extract joint coordinate information from video frames. Our spatio-temporal transformer encoder model is enhanced with dual attention and weighted residual connections for assessing human motion quality using skeleton data. Our model has been evaluated on a custom RGB dataset, as well as two benchmark datasets—UI-PRMD and KIMORE, and achieves state-of-the-art performance in motion quality assessment, demonstrating an average reduction of 12.9% in mean absolute error (MAE) across all datasets. Our approach has the potential to be a cost-effective and reliable solution for home-based rehabilitation, offering advanced motion assessment capabilities without the need for expensive equipment.</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

Towards an RGB camera-based rehabilitation exercise assessment using an enhanced spatio-temporal transformer framework

  • 期刊论文

  • 发表日期：

    2025年8月6日

查看全文
  • Amanpreet Chander,
  • Deepti R. Bathula,
  • Ashish Kumar Sahani

摘要

Rehabilitation following surgery or trauma is crucial for patient recovery. However, many patients attempt to rehabilitate at home by performing exercises without professional supervision, potentially hindering their progress or causing complications. To address this, deep learning systems have been extensively explored. Traditional motion detection systems such as Vicon or Kinect, while effective, are costly or impractical for at-home use. Therefore, leveraging the ubiquity of mobile devices, we propose an affordable and robust home-based rehabilitation system that uses a simple RGB camera for recording physical exercises. We used Mediapipe, an open-source framework, to extract joint coordinate information from video frames. Our spatio-temporal transformer encoder model is enhanced with dual attention and weighted residual connections for assessing human motion quality using skeleton data. Our model has been evaluated on a custom RGB dataset, as well as two benchmark datasets—UI-PRMD and KIMORE, and achieves state-of-the-art performance in motion quality assessment, demonstrating an average reduction of 12.9% in mean absolute error (MAE) across all datasets. Our approach has the potential to be a cost-effective and reliable solution for home-based rehabilitation, offering advanced motion assessment capabilities without the need for expensive equipment.