End-to-End Deep Learning for Stress Recognition Using Remote Photoplethysmography

With the development of remote Photoplethysmography, contactless measurement of physiological parameters is possible. Several works have investigated stress recognition based on heart rate variability features extracted from cameras. In this paper, we study the application of deep learning networks for end-to-end cognitive stress recognition using remote Photoplethysmography. We compared different network architectures in an experiment with 15 participants at rest and under cognitive stress. The experimental results show that the CNN-based networks can learn stress-related characteristics even from inter-beat-interval signals of short length. The deep network-based approaches outperformed the classical HRV feature-based methods.