RePhys: Lightweight Heart Rate Measurement Network for Face Videos Based on Reparameterization

The widespread use of smart mobile devices in modern society presents significant potential for the extensive integration of rPPG technology on mobile platforms. Our primary focus is on designing a lightweight network, RePhys, for measuring heart rate signals based on reparameterization. This involv...

Full description

Saved in:
Bibliographic Details
Main Authors: Nie, Weiqi, Tao, Kemeng, Chen, Jianzhang, Liu, Zhi, Miao, Danmin
Format: Conference Proceeding
Language:English
Subjects:
Accuracy
deep learning
Heart rate detection
Heart rate measurement
Learning (artificial intelligence)
lightweight network
Mobile handsets
Performance evaluation
physiological signal measurement
reparameterization
rPPG
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The widespread use of smart mobile devices in modern society presents significant potential for the extensive integration of rPPG technology on mobile platforms. Our primary focus is on designing a lightweight network, RePhys, for measuring heart rate signals based on reparameterization. This involves streamlining network parameters and floating-point operations to enhance performance and operational efficiency on mobile devices while maintaining measurement accuracy. RePhys was trained using two public datasets, UBFC and MMPD, and its training effectiveness was assessed. Experimental results reveal that RePhys achieved the highest accuracy across all indicators on the UBFC dataset, with competitive performance on MMPD. Ablation experiments were subsequently conducted to validate the efficacy of each component of the RePhys network. Finally, a comparative analysis was undertaken, assessing parameter quantities, floating-point operations, and delay times of RePhys against various existing networks, highlighting the superior lightweight design of RePhys.
ISSN:2769-3554
DOI:10.1109/ICAIBD62003.2024.10604535