Generisch-Net: A Generic Deep Model for Analyzing Human Motion with Wearable Sensors in the Internet of Health Things

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Human-Computer Interaction

Background:

• The Internet of Health Things (IoHT) aims for autonomous, low-cost healthcare interventions.
• Wearable Inertial Measurement Units (IMUs) offer insights into human motion, psychological state, and daily activities.
• Current research often uses specialized deep models for specific human motion analysis tasks.

Purpose of the Study:

• To develop a single, versatile deep-learning model for multiple human motion analysis applications.
• To demonstrate the efficacy of a generic BiGRU-CNN model for predicting emotional state, classifying activities of daily living, and re-identifying individuals.
• To leverage wearable IMU data for comprehensive human motion analysis.

Main Methods:

• Development of a generic Bidirectional Gated Recurrent Unit-Convolutional Neural Network (BiGRU-CNN) model.
• Utilized both publicly available and closed-access datasets for training and validation.
• Collected data non-invasively using wearable IMUs attached to subjects.

Main Results:

• The generic model achieved 96.97% accuracy in classifying activities of daily living.
• Individual re-identification in closed-loop scenarios reached 93.71% accuracy.
• Emotional state estimation accuracy was 78.20%.

Conclusions:

• A versatile deep-learning model using wearable IMUs shows significant promise for diverse human motion analysis applications.
• The generic BiGRU-CNN model offers a unified approach, reducing the need for problem-specific models.
• This study advances the development of intelligent systems within the Internet of Health Things framework.