Sleep/wake classification using remote PPG signals

Abstract

This paper proposes a remote sleep/wake classification method by combining vision-based heart rate (HR) estimation and convolutional neural network (CNN). Instead of directly inputting the estimated HR to CNN, we input remote PPG (Photoplethysmogram) signals filtered by a dynamic HR filter, which can overcome two main problems: low temporal resolution of estimated HR; much noise exists in the estimated remote PPG signals. Evaluation results show that the dynamic HR filter works more effectively compared to the static one, which helps improve AUC (area under the curve) index of the classification to 0.70, as good as the performance (0.71) of HR from a wearable sensor.