Skeleton-Based Posture Classification to Promote Safer Walker-Assisted Gait in Older Adults

2026-05-05 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Medical Devices & Health Technology · Depth: Advanced, extended

Summary

This study evaluates machine learning models for skeleton-based posture classification to enhance smart walkers for older adults, aiming to prevent falls. Researchers compared a Geometric approach, XGBoost, SVM, and deep learning architectures (4-layer CNN, 6-layer CNN, Encoder-Decoder CNN) for classifying walker usage, standing vs. sitting, and specific postures. Data was collected from 21 participants (ages 21-48) using a Socially Assistive Walker (SAW) equipped with a Raspberry Pi 4, USB cameras, LiDAR, and force sensors, leveraging Google MediaPipe for 3D pose landmark detection. XGBoost and the Geometric approach emerged as top performers, with XGBoost achieving near-perfect training accuracy (99.84% for walker choice, 99.69% for standing vs. sitting) and 85% prediction accuracy for "bad posture" detection. The Geometric approach attained 89.9% overall accuracy for 8 postures. Deep learning models also showed strong performance in binary classification tasks, exceeding 98% accuracy.

Key takeaway

For Computer Vision Engineers developing assistive technologies for older adults, this research indicates that integrating marker-less pose classification into smart walkers can significantly improve fall prevention. You should prioritize robust models like XGBoost or a Geometric approach for real-time posture detection, as they demonstrated superior performance and generalization compared to more complex deep learning models in this context. Focus on binary classifications (e.g., standing/sitting, walker use) where models achieve near-perfect accuracy, and consider combining geometric methods with sensor data for specific "bad posture" detection.

Key insights

Machine learning models can effectively classify posture and walker use to enhance fall prevention in smart walkers.

Principles

• Marker-less tracking improves user comfort.
• Simpler models can outperform complex deep learning.
• Personalized systems enhance detection accuracy.

Method

The method involves acquiring full-body and upper-body pose landmarks via walker-mounted cameras, extracting features like joint distances and angles, and then training various supervised learning models for binary (walker use, standing/sitting) and multi-class (17 postures) classification.

In practice

• Use Google MediaPipe for real-time pose detection.
• Consider XGBoost for robust posture classification.
• Integrate force sensors for enhanced hand posture detection.

Topics

• Skeleton-Based Posture Classification
• Smart Walkers
• Fall Prevention
• XGBoost Algorithm
• Deep Learning Architectures

Best for: Computer Vision Engineer, AI Scientist, Robotics Engineer, Research Scientist

Related on AIssential

• A machine learning study highlighting the challenges of fidgety movement recognition using vision and inertial sensors — Automated Fidgety Movement recognition in infants faces significant generalization challenges across unseen subjects.
• An interpretable and trustworthy AI framework for large-scale longitudinal structure-pain association studies using data from the Osteoarthritis Initiative (OAI) — Combining uncertainty-aware deep learning with longitudinal statistical modeling improves osteoarthritis structure-pain association studies.
• Quality Adaptive Angular Margin Learning for Respiratory Sound Classification — QLung uses quality-adaptive angular margins to improve respiratory sound classification generalization and handle class imbalance.
• Generative Data Augmentation for Skeleton Action Recognition — A conditional generative pipeline synthesizes diverse skeleton data to augment action recognition, improving performance in low-data settings.
• Kaggle Solution Walkthroughs: Enefit - Predict Energy Behavior of Prosumers with HYD — Combining XGBoost and GRU with online learning and monthly targets significantly improves model performance.
• Explainable Fall Detection for Elderly Care via Temporally Stable SHAP in Skeleton-Based Human Activity Recognition — Temporally stable SHAP explanations enhance reliability for skeleton-based fall detection in elderly care.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.