Hidden Markov Model-Based Fall Detection With Motion Sensor Orientation Calibration: A Case for Real-Life Home...

Area of Science:

• Gerontology
• Biomedical Engineering
• Signal Processing

Background:

• Falls pose a significant risk to the independence of senior citizens.
• Motion sensor technology offers a low-cost solution for monitoring and fall detection.
• Existing systems often struggle with real-world complexities like sensor misplacement.

Purpose of the Study:

• To develop an automatic fall detection system using a hidden Markov model (HMM) and a single motion sensor.
• To address challenges of feature engineering and sensor orientation issues in real-life monitoring.
• To evaluate the system's performance in simulated and real-world fall scenarios.

Main Methods:

• Development of a hidden Markov model (HMM) based fall detection system.
• Introduction of a novel acceleration signal representation for HMMs, eliminating manual feature engineering.
• Implementation of a sensor orientation calibration algorithm to correct for misplacement.
• Training HMM classifiers on acceleration data from simulated and real-world fall datasets.

Main Results:

• The system achieved high accuracy on an experimental dataset (PPV: 0.981, Sensitivity: 0.992) with simulated falls.
• Performance on a real-world fall dataset (FARSEEING) showed strong results (PPV: 0.786, Sensitivity: 1.000).
• The HMM-based system with calibration significantly outperformed benchmark systems.

Conclusions:

• The developed HMM-based fall detection system effectively detects falls in real-life home environments.
• Sensor orientation calibration enhances system robustness against real-world deployment issues.
• The system offers precise fall detection with a reasonably low false alarm rate, supporting independent living for seniors.