HR-TRACK: An rPPG Method for Heartrate Monitoring Using Temporal Convolution Networks

Lokendra Birla, Sneha Shukla, Trishna Saikia, Puneet Gupta

Abstract: The COVID-19 pandemic necessitates avoiding skin contact to minimize the spread of virus infection. It paves the way for an active surge in telehealthcare research. In this direction, Remote Photoplethysmography (rPPG) plays a crucial role in analyzing heart rate (HR) from non-contact face videos. Existing rPPG-based HR monitoring methods fail when face video duration is small and the video contains facial deformations. These issues are mitigated by our proposed method HR-TRACK, that is, rPPG method for Heart Rate moniToring using tempoRAl Convolution networK. It improves HR monitoring by introducing a novel architecture formed by sequentially stacking two novel networks. The networks are inspired by the temporal convolution network (TCN) to model long temporal sequences effectively. Our first network automatically mitigates the noise induced by facial deformations and performs blind source separation to predict pulse signals. The instantaneous HR obtained from the pulse signal can be erroneous. Thus, our second network analyzes all the computed HR values and rectifies the erroneous HR, if any. The experimental results conducted on the publicly available datasets reveal that our proposed method outperforms the state-of-the-art methods. Furthermore, the results justify the utilization of both networks to improve HR monitoring.

External IDs:dblp:conf/icpr/BirlaSSG24