Seamless tracing of human behavior using complementary wearable and house-embedded sensors

Area of Science:

• Gerontology
• Computer Science
• Biomedical Engineering

Background:

• Elderly surveillance systems require unobtrusive and accurate monitoring.
• Multimodal sensor fusion offers potential for improved data reliability and comprehensiveness.

Purpose of the Study:

• To develop and evaluate a multimodal surveillance system for elderly individuals.
• To assess the effectiveness of combining visual and accelerometer-based mobility sensors.
• To implement an automatic danger detection algorithm.

Main Methods:

• Utilized a multimodal approach with wearable (accelerometer) and premise-embedded (visual) sensors.
• Employed polar histogram-based visual pose recognition and dynamic time warping for action analysis.
• Developed an automatic danger detection algorithm using premise- and subject-related databases.

Main Results:

• Achieved 95.5% accuracy for elementary pose recognition using video and 96.7% using accelerometers.
• Combined accelerometer and video systems reached 98.9% accuracy for elementary pose recognition.
• Demonstrated 80% accuracy for complex outdoor activity recognition using the accelerometer system.

Conclusions:

• Multimodal sensor fusion significantly enhances the accuracy of elderly pose recognition.
• The developed system offers a robust solution for seamless surveillance and automatic danger detection.
• Real-life data integration and advanced algorithms contribute to the system's novelty and effectiveness.