SEQuence-rPPG: A Fast BVP Signal Extraction Method From Frame Sequences

Kegang Wang, Yantao Wei, Mingwen Tong, Jie Gao, ZhongJin Zhao, YuJian Ma, Yi Tian

Keywords: rPPG, Remote vital sensing, Signal processing

TL;DR: A new rPPG method is proposed, which is very simple, fast and accurate.

Abstract: Non-contact heart rate estimation has essential implications for the development of affective computing and telemedicine. However, existing deep learning-based methods often endeavor to achieve real-time measurements, so a simple, fast, pre-processing-free approach is needed. Our work consists of two main parts. Firstly, we proposed SEQ-rPPG, which first transforms the RGB frame sequence into the original BVP signal sequence by learning-based linear mapping and then outputs the final BVP signal using 1DCNN-based spectral transform, and time-domain filtering. Secondly, to address the shortcomings of the existing dataset in training the model, a new large-scale dataset was collected for training and testing. Our approach achieved competitive results on the collected large dataset(the best) and public dataset UBFC-rPPG(0.81 MAE with 30s time window, test only). It requires no complex pre-processing, has the fastest speed, can run in real-time on mobile ARM CPUs, and can achieve real-time beat-to-beat performance on desktop CPUs. Benefiting from the high-quality training set, other deep learning-based models reduced errors by at least 53$\%$. We compared the methods with and without the spectral transformation, and the results show that the processing in the time domain is effective.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Machine Learning for Sciences (eg biology, physics, health sciences, social sciences, climate/sustainability )