Radar-Based Elderly Fall Detection Using Smoothed Pseudo Wigner Ville Distribution and XGBoost Learning.

Area of Science:

• Gerontology
• Biomedical Engineering
• Signal Processing

Background:

• Falls in the elderly are a major health concern, leading to significant morbidity and reduced quality of life.
• Traditional fall detection systems, like wearables and cameras, face limitations including privacy issues and environmental dependencies.
• Radar-based systems offer an unobtrusive and privacy-preserving approach to fall detection.

Purpose of the Study:

• To develop and evaluate a novel radar-based fall detection system for the elderly.
• To classify fall events using smoothed pseudo Wigner-Ville distribution (SPWVD) images and XGBoost machine learning.
• To assess the accuracy and efficiency of the proposed method in distinguishing falls from non-fall events.

Main Methods:

• Utilized an online publicly available radar database (N=15) comprising radar signals.
• Applied smoothed pseudo Wigner-Ville distribution (SPWVD) to radar signals for time-frequency representation images.
• Extracted ten features from SPWVD images and applied them to XGBoost learning, evaluated using 10-fold cross-validation.

Main Results:

• The proposed radar-based approach achieved high performance metrics, including a maximum average classification accuracy of 87.47%.
• Key performance indicators demonstrated the system's effectiveness: f1-score (87.38%), precision (88.12%), sensitivity (86.81%), specificity (88.31%), and kappa score (74.94%).
• A combination of conventional features with concentration measures and median frequency yielded the second-best performance.

Conclusions:

• The developed framework demonstrates accurate and efficient detection of falls among the elderly population.
• Radar-based fall detection using SPWVD images and XGBoost is a viable and promising alternative to existing methods.
• The system's unobtrusive nature makes it suitable for deployment in private living spaces, enhancing elderly safety and independence.