Scholarlink

Contemporary society is gradually aging, accentuating the increasing importance of ensuring the health and safety of the elderly in the healthcare and public health sectors. In particular, the fall risks of the elderly become a significant concern. To address this, the OpenPose human posture recognition technology is integrated with interpretable machine learning models to develop a fall risk assessment framework based on the key parts of the human body. Given the significance of human movement pattern analysis in assessing health conditions and rehabilitation progress, we employed exploratory data analysis. Based on motion data collected from 63 patients, we explored the variations in joint movements at distinct stages (first 15 s and last 15 s). Initially, we examined the data distribution of each joint along the x and y directions and observed notable variations in areas such as the wrists and nose. By computing the correlation coefficients between joint points, we identified higher correlations between upper limb joints. We employed principal component analysis to reduce the dimensionality of the data. The results revealed temporal partition trends along the x direction. Subsequently, k-means clustering showed the distinct separation of sample points across different times. By exploring the variation patterns evident in human joint data through various data analysis methods, distinctions between different stages were confirmed, and key joints with significant influence were identified. The results provide a basis for further research on mechanisms governing joint motion.

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

Interpretable Machine Learning for Fall Risk Assessment: An Exploratory Study Using OpenPose and Key Point Motion Data

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年

查看全文
  • Chih-Ching Chang,
  • Chia-Hsuan Lee,
  • Ying-Po Hsu,
  • Tien-Lung Sun

摘要

Contemporary society is gradually aging, accentuating the increasing importance of ensuring the health and safety of the elderly in the healthcare and public health sectors. In particular, the fall risks of the elderly become a significant concern. To address this, the OpenPose human posture recognition technology is integrated with interpretable machine learning models to develop a fall risk assessment framework based on the key parts of the human body. Given the significance of human movement pattern analysis in assessing health conditions and rehabilitation progress, we employed exploratory data analysis. Based on motion data collected from 63 patients, we explored the variations in joint movements at distinct stages (first 15 s and last 15 s). Initially, we examined the data distribution of each joint along the x and y directions and observed notable variations in areas such as the wrists and nose. By computing the correlation coefficients between joint points, we identified higher correlations between upper limb joints. We employed principal component analysis to reduce the dimensionality of the data. The results revealed temporal partition trends along the x direction. Subsequently, k-means clustering showed the distinct separation of sample points across different times. By exploring the variation patterns evident in human joint data through various data analysis methods, distinctions between different stages were confirmed, and key joints with significant influence were identified. The results provide a basis for further research on mechanisms governing joint motion.