Uni-rPPGNet: Efficient and Lightweight Remote Heart Rate Variability Measurement

Heart rate variability (HRV), a critical physiological parameter indicative of an individual's physiological state, stands as one of the primary monitoring indices within the medical field. In recent years, low-cost non-contact measurement technologies have garnered extensive attention. Nonetheless, due to the complexities involved in measuring HRV, there are scant methods available for remote HRV measurement. This paper introduces an efficient end-to-end rPPG signal extraction network architecture based on 3DCNN and Transformer—Uni-rPPGNet. The network employs the Stem system to extract coarse features, thereby reducing input tensor dimensions. It utilizes shallow 3DCNN to derive local features and incorporates a 3D ShuffleAttention module to facilitate feature communication across channels. Deeper layers engage an hourglass-shaped Transformer to capture global context features. Through training and evaluation on the UBFC-rPPG and PURE datasets, the results reveal that Uni-rPPGNet surpasses other existing methods in measuring HRV indices, demonstrating the technology's effectiveness and utility by offering a novel approach for remote HRV measurement.