A Combined One-Class SVM and Template-Matching Approach for User-Aided Human Fall Detection by Means of Floor...

Area of Science:

• Gerontology
• Biomedical Engineering
• Acoustic Signal Processing

Background:

• Falls are the leading cause of accidental death and serious injury among older adults.
• Early detection of falls is crucial for timely medical intervention and reducing mortality.
• Existing acoustic-based fall detection systems require further improvement in accuracy and robustness.

Purpose of the Study:

• To develop and evaluate a novel, user-aided, two-stage acoustic fall detection system for the elderly.
• To improve the accuracy and reliability of fall detection using a combination of machine learning classifiers.
• To reduce false positives and enhance the practical applicability of acoustic fall detection.

Main Methods:

• A single Floor Acoustic Sensor captures audio signals.
• Mel-Frequency Cepstral Coefficients (MFCCs) and Gaussian Mean Supervectors (GMSs) are extracted for feature representation.
• A semi-supervised framework combining One-Class SVM (OCSVM) for anomaly detection and template-matching for classification is employed.

Main Results:

• The proposed OCSVM and template-matching system demonstrated improved performance over OCSVM-only methods.
• Performance gains of 10.14% in clean and 4.84% in noisy conditions were observed compared to OCSVM alone.
• Significant improvements of 19.96% in clean and 8.08% in noisy conditions were achieved compared to prior work (Popescu and Mahnot, 2009).

Conclusions:

• The proposed two-stage, user-aided acoustic fall detection system offers a promising solution for elder safety.
• The combined classifier approach effectively discriminates falls from non-falls, even in challenging acoustic environments.
• This method represents a significant advancement in acoustic-based fall detection technology for the elderly population.