Fusing IoT Wearable and Object Motion Sensors for Enhanced Activity Recognition in Smart Homes

Area of Science:

• Gerontology and Health Informatics
• Human-Computer Interaction
• Machine Learning Applications in Healthcare

Background:

• Increasing elderly population necessitates technology for independent living and aging in place.
• Shortage of aged care staff highlights the need for remote monitoring solutions.
• Internet of Things (IoT) offers potential for daily activity monitoring and timely alerts.

Purpose of the Study:

• To investigate the integration of wearable and object-based IoT motion sensors for accurate human activity recognition.
• To assess the independent living ability of older adults through sensor-based activity monitoring.
• To enhance accuracy, scalability, and cost-effectiveness of activity recognition systems.

Main Methods:

• Collected real-world data from 17 key daily activities using 30 wearable and object motion sensors.
• Implemented and compared machine learning algorithms for activity classification.
• Optimized sensor configuration for improved scalability and cost-efficiency.

Main Results:

• Random Forest (RF) algorithm achieved 90.63% accuracy with 11 wearable sensors.
• Accuracy improved to 97.88% when 19 object sensors were combined with wearable sensors.
• Reduced sensor count by 57% (from 30 to 13) without compromising performance.

Conclusions:

• Integrated IoT motion sensors provide a robust mechanism for monitoring daily activities of older adults.
• The system enables real-time alerts for caregivers and clinicians, supporting timely interventions.
• This technology promotes the well-being and independence of older adults, facilitating aging in place.