Elderly Fall Detection Based on GCN-LSTM Multi-Task Learning Using Nursing Aids Integrated with Multi-Array Flexible...

Area of Science:

• Biomedical Engineering
• Gerontology
• Sensor Technology

Background:

• Elderly falls pose significant health risks, necessitating effective detection methods.
• Current fall detection systems often rely on costly and complex visual or multi-sensor devices.
• Limited applicability of existing methods due to cost and design complexity.

Purpose of the Study:

• To develop a cost-effective and widely applicable fall detection method for the elderly.
• To propose a fall detection system utilizing nursing aids with integrated multi-array flexible tactile sensors.
• To leverage plantar force analysis and tactile data for accurate fall detection.

Main Methods:

• Design and implementation of multi-array capacitive tactile sensors.
• Distribution of sensors on the foot based on plantar force analysis.
• Development of a fall detection model using a graph convolution neural network (GCN) and long-short term memory (LSTM) network (GCN-LSTM).
• Extraction of spatial and temporal features from tactile sequences using GCN and LSTM modules.

Main Results:

• Achieved a Mean Squared Error (MSE) of 0.0716 for predicted tactile data.
• Demonstrated a fall detection accuracy of 96.36%.
• Successfully performed future fall detection up to 5 time steps (0.2-s intervals) with high confidence.
• Exhibited robust generalization capabilities across different ground types and morphologies.

Conclusions:

• The proposed GCN-LSTM model effectively detects falls using tactile data from nursing aids.
• This tactile sensor-based approach offers a promising, accurate, and robust alternative to existing fall detection systems.
• The method shows potential for widespread adoption in elderly care due to its cost-effectiveness and design simplicity.