Contactless Fall Detection by Means of Multiple Bioradars and Transfer Learning

Area of Science:

• Geriatric care
• Biomedical engineering
• Signal processing

Background:

• Automatic fall detection is crucial for elderly individuals living alone to prevent serious health consequences.
• Contactless fall detection using bioradiolocation shows promise but requires improved precision and view-independent recognition.
• Existing systems can be enhanced for better accuracy in recognizing falls.

Purpose of the Study:

• To develop and evaluate a more precise, contactless fall detection system for the elderly.
• To investigate the effectiveness of multi-channel bioradar systems combined with advanced signal processing and machine learning.
• To improve upon existing two-bioradar systems for fall recognition.

Main Methods:

• A multi-channel bioradar system (three or four bioradars) was employed for data acquisition.
• Wavelet transform was utilized for signal processing.
• A pre-trained convolutional neural network (AlexNet) was fine-tuned using scalograms for binary classification of falls.
• Experiments involved various radar configurations to record different movement types.

Main Results:

• The four-bioradar system achieved an accuracy of 0.99 and a Cohen's kappa of 0.99.
• The three-bioradar system demonstrated high performance with an accuracy of 0.98 and Cohen's kappa of 0.97.
• Both multi-bioradar systems significantly outperformed the traditional two-bioradar system (accuracy 0.92, Cohen's kappa 0.86).

Conclusions:

• A multi-channel bioradar system, utilizing wavelet transform and transfer learning, offers a significant improvement in contactless fall detection accuracy.
• The proposed system holds potential for the development of advanced, reliable fall detection solutions for the elderly.
• Further research can explore optimizing radar configurations and deep learning models for enhanced geriatric safety.