ECG-based Daily Activity Recognition Using 1D Convolutional Neural Networks.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Wearable Technology

Background:

• Electrocardiogram (ECG) signals traditionally monitor cardiac health.
• Expanding ECG applications for broader patient surveillance is an emerging area.
• Human Activity Recognition (HAR) systems often rely on other sensor types.

Purpose of the Study:

• To develop and validate a human activity recognition (HAR) system using only electrocardiogram (ECG) signals.
• To explore the potential of ECG for comprehensive patient monitoring beyond cardiophysiological data.
• To address limitations of previous HAR studies using smaller, public datasets.

Main Methods:

• An end-to-end one-dimensional convolutional neural network (1D CNN) model was utilized.
• Wireless ECG data were collected from 40 participants engaged in five common daily activities.
• A subject-independent methodology with 5-fold cross-validation was implemented for robust evaluation.

Main Results:

• The HAR system achieved a test accuracy of 82.9% across all activities.
• The model demonstrated high efficacy in recognizing specific activities, notably 'Sleeping' with 98.5% accuracy.
• The study confirmed the generalizability and applicability of the developed HAR system.

Conclusions:

• ECG signals are a practical and effective data source for human activity recognition (HAR).
• This approach enables advanced patient surveillance, including emergency detection, extending beyond cardiac monitoring.
• The findings support the integration of ECG-based HAR into comprehensive healthcare solutions.