HMM-Based Action Recognition System for Elderly Healthcare by Colorizing Depth Map

Area of Science:

• Computer Vision
• Artificial Intelligence
• Gerontology

Background:

• Elderly care requires reliable monitoring systems, but current action recognition technology is not suitable for continuous, automated use.
• Existing methods often compromise privacy or lack efficiency in real-world applications.

Purpose of the Study:

• To develop and validate a novel, real-time action recognition system for elderly care applications.
• To enhance privacy protection and operational efficiency in continuous monitoring.

Main Methods:

• A real-time action recognition system integrating Hidden Markov Models (HMM) with colorized depth maps.
• Privacy-preserving person detection using You Only Look Once (YOLOv5) on depth data.
• Feature extraction via Histogram of Oriented Gradients (HOG) from depth map sequences, processed by HMM and Viterbi Algorithm for action recognition.

Main Results:

• The system demonstrated effective action recognition on real-world data from three participants in a care center.
• Fusion of HMM with Support Vector Machine (SVM) yielded the highest average accuracy of 84.04% among tested classification algorithms.

Conclusions:

• The developed system provides a robust and privacy-conscious solution for real-time action recognition in elderly care settings.
• The HMM-SVM fusion approach shows significant promise for improving automated monitoring and support for the elderly.