Recognition of everyday activities using experiment data from wearable sensors: a deep learning-based framework

Area of Science:

• Gerontology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Continuous monitoring of older adults' daily activities is crucial for early health change detection and timely intervention.
• Wearable sensors and deep learning offer promising solutions for unobtrusive health monitoring.
• Existing methods often require numerous sensors, limiting practical application.

Purpose of the Study:

• To develop an efficient method for recognizing everyday activities in older adults using minimal wearable sensors and deep learning.
• To evaluate the impact of sensor count and placement on activity recognition accuracy.
• To identify an optimal sensor configuration for reliable activity monitoring.

Main Methods:

• A small-scale home laboratory experiment was conducted to recognize 14 distinct everyday activities.
• Five deep learning models were compared based on sensor signal count and recognition accuracy.
• Activity recognition performance was assessed using varying numbers of wearable sensors placed on the body.

Main Results:

• Sensor placement significantly impacts activity recognition accuracy.
• A two-sensor system (pelvis and right hand) achieved 89.3% accuracy in recognizing 12 activities.
• A five-sensor system recognized 14 activities with 88.2% accuracy, while a single-sensor approach showed poor performance.
• The two-sensor system demonstrates potential for large-scale data collection.

Conclusions:

• A minimal two-sensor configuration (pelvis and right hand) is effective for accurate everyday activity recognition in older adults.
• This optimized sensor approach balances accuracy and efficiency for practical health monitoring.
• The developed method facilitates large-scale data collection for enhanced understanding of older adults' daily lives and health status.