Unified framework for triaxial accelerometer-based fall event detection and classification using cumulants and...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Falls are a significant health risk, especially for the elderly.
• Accurate fall detection systems are crucial for timely intervention.
• Existing methods often require multiple sensors or complex feature extraction.

Purpose of the Study:

• To develop a unified framework for fall event detection and classification.
• To identify optimal cumulant features and classifiers for analyzing accelerometer signals.
• To achieve high accuracy and low false alarm rates in fall detection.

Main Methods:

• Utilized cumulants (second- and fifth-order) extracted from triaxial accelerometer (ACC) signals.
• Developed a hierarchical decision tree algorithm incorporating Support Vector Machine (SVM) classifiers.
• Compared performance against other classifiers (Decision Tree, Naive Bayes, Multilayer Perceptron) and time-domain features.

Main Results:

• The proposed framework achieved detection and classification rates exceeding 95%.
• Fifth-order cumulants and SVM were effective for both fall detection and classification.
• Second-order cumulants and SVM enabled accurate human activity classification.
• The system demonstrated a minimal false alarm rate of 1.03%.

Conclusions:

• A unified framework using cumulants and SVM provides an effective solution for fall detection and classification.
• Second- and fifth-order cumulants are robust features for analyzing accelerometer data in human activity recognition.
• The proposed method offers high accuracy and reliability for wearable fall monitoring systems.