Acoustic fall detection using one-class classifiers.

Area of Science:

• Gerontology
• Biomedical Engineering
• Signal Processing

Background:

• Falls pose a significant health risk to the elderly population.
• Existing acoustic fall detection systems require extensive training data, including realistic fall sounds, which are difficult to obtain.
• The FADE system aims to alert caregivers to falls, supporting elderly individuals living alone.

Purpose of the Study:

• To investigate the efficacy of one-class classifiers for acoustic fall detection.
• To address the challenge of limited realistic fall sound data for training classifiers.
• To compare the performance of one-class classifiers against traditional two-class classifiers.

Main Methods:

• Utilized three one-class (OC) classifiers: nearest neighbor (OCNN), SVM (OCSVM), and Gaussian mixture (OCGM).
• Trained classifiers using only non-fall sound events.
• Evaluated classifier performance on two distinct datasets, comparing OC methods to regular two-class classifiers.

Main Results:

• One-class classifiers demonstrated potential in distinguishing falls from daily activities without needing fall sound examples.
• Performance metrics indicated the viability of OC approaches in this challenging domain.
• Comparison revealed insights into the relative strengths of OCNN, OCSVM, and OCGM for fall detection.

Conclusions:

• One-class classification offers a promising solution for acoustic fall detection systems, overcoming data acquisition limitations.
• The FADE system, enhanced with OC classifiers, can improve the safety and independence of elderly individuals.
• Further research and validation are warranted to optimize OC classifier implementation in real-world fall detection applications.