Autoregressive-moving-average hidden Markov model for vision-based fall prediction-An application for walker robot.

Area of Science:

• Robotics and Human-Computer Interaction
• Geriatric Technology and Assistive Devices
• Machine Learning for Healthcare

Background:

• Population aging necessitates advanced assistive technologies for elderly safety.
• Fall detection and prediction are critical for preventing injuries in older adults.
• Existing fall detection methods require improvement for real-time, proactive intervention.

Purpose of the Study:

• To develop and evaluate a real-time fall prediction algorithm for users of walking assistive systems.
• To enhance the safety and independence of the elderly through advanced technology.
• To leverage visual data from depth sensors for predictive fall analysis.

Main Methods:

• A hybrid system identification and machine learning approach was employed.
• An Autoregressive-Moving-Average (ARMA) model forecasted future walking states from time-series data.
• A Hidden Markov Model (HMM) classifier, built upon the ARMA model, predicted falls.

Main Results:

• The algorithm achieved an 84.72% success rate in predicting falls across various scenarios.
• Experiments included diverse falling scenarios (forward, down, back, left, right).
• Algorithm performance was validated using data from four subjects, including a physiotherapist.

Conclusions:

• The proposed real-time fall prediction algorithm shows significant promise for enhancing elderly safety.
• This technology can provide timely alerts, potentially preventing falls and related injuries.
• The hybrid ARMA-HMM approach offers a robust method for proactive fall risk assessment in assistive systems.