Intelligent ADL Recognition via IoT-Based Multimodal Deep Learning Framework

Area of Science:

• Computer Science
• Gerontology
• Biomedical Engineering

Background:

• Remote monitoring systems are crucial for elder care, offering families and caregivers flexibility.
• Activities of Daily Living (ADLs) provide an effective metric for monitoring elderly individuals and patients.
• Existing systems often rely on single-type sensors, limiting comprehensive monitoring.

Purpose of the Study:

• To propose a robust, layered architecture for remote elderly monitoring using multisensory Internet of Things (IoT) devices.
• To develop a multimodal approach integrating wearable sensors and video data for enhanced ADL recognition.
• To achieve accurate and reliable remote monitoring of elderly individuals' daily activities.

Main Methods:

• A layered architecture processing data from multimodal IoT sensors (wearable inertial sensors, video).
• Pre-processing steps included data filtration, segmentation, landmark detection, and 2D stick model creation.
• Feature extraction, fusion, and optimization were performed, followed by classification using a Convolutional Neural Network (CNN).

Main Results:

• The proposed multimodal system effectively fused data from diverse sensors.
• The layered architecture successfully processed and analyzed sensor data for ADL recognition.
• An acceptable mean accuracy rate of 84.14% was achieved in recognizing activities of daily living.

Conclusions:

• The developed IoT-based multimodal layered system demonstrates a viable solution for remote elderly monitoring.
• The integration of multisensory data and deep learning enhances the accuracy of ADL recognition.
• This approach offers a promising direction for improving the safety and well-being of elderly individuals at home.