Machine Learning-Based Classification of Human Behaviors and Falls in Restroom via Dual Doppler Radar Measurements

Area of Science:

• Engineering
• Computer Science
• Gerontology

Background:

• Traditional restroom monitoring systems raise privacy concerns.
• Accurate fall detection is crucial for elderly care and independent living.
• Existing remote sensing technologies have limitations in restroom environments.

Purpose of the Study:

• To develop and validate a novel radar-based system for non-invasive human behavior and fall classification in restrooms.
• To compare the performance of various machine learning algorithms for analyzing radar data.
• To identify key features for accurate behavior recognition in this specific environment.

Main Methods:

• Deployment of a dual Doppler radar system on the ceiling and wall of a restroom.
• Application of machine learning algorithms: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Support Vector Machine (SVM), and Random Forest.
• Utilizing Doppler spectrograms (time-velocity distribution) as input features for classification.
• Experimental validation with 21 participants performing 8 distinct behaviors, including falls.

Main Results:

• The CNN model achieved the highest overall classification accuracy of 95.6% and 100% accuracy for fall detection.
• The radar system demonstrated superior performance compared to conventional thermal sensors and other radar techniques.
• Analysis revealed that higher-order derivatives of acceleration and jerk, along with horizontal motion information, are effective features for behavior classification.

Conclusions:

• The proposed radar system offers an accurate and privacy-conscious solution for monitoring human behaviors and detecting falls in restrooms.
• Machine learning, particularly CNNs, effectively processes Doppler radar data for high-accuracy classification.
• This technology holds significant potential for enhancing safety and enabling timely assistance in domestic and care settings.