Remote Heart Rate Estimation using Swin Transformer V2 and Wrapping Temporal Shift Modules

Abstract

Remote heart rate (rHR) estimation, which aims to measure heart activities without any physical contact with the subject, is performed using remote photoplethys- mography (rPPG) and has great potential in many applications. In this paper, we introduce a remote heart rate (rHR) estimation algorithm using Swin Transformer V2 and Wrapping Temporal Shift Modules (WTSM). Moreover, we apply difference layer to reduce the lighting and motion noise and shallow stem to extract coarse local spatio-temporal features. Finally, we apply linear layer to project features to 1D rPPG signal and estimate heart rate using FFT. To evaluate the performance of the proposed algorithm, we train and test on the public UBFC-rPPG and PURE dataset. The experimental results show that the proposed algorithm achieve better accuracy than CNN based methods.