Search papers, labs, and topics across Lattice.
This paper introduces StreamPPG, a novel architecture for low-latency remote photoplethysmography (rPPG) that estimates blood volume pulse signals from facial videos in real-time. By employing a consistent privileged learning (CPL) strategy, StreamPPG effectively utilizes ground-truth rPPG signals to enhance model representation, addressing the limitations of both conventional clip-wise and frame-wise methods. Experimental results show that StreamPPG achieves state-of-the-art accuracy while maintaining the necessary throughput for real-time applications on edge devices.
StreamPPG achieves real-time rPPG estimation with state-of-the-art accuracy by leveraging privileged information, overcoming the latency barriers of traditional methods.
Remote photoplethysmography (rPPG) estimates the blood volume pulse (BVP) signal from facial videos, enabling contact-free health monitoring. Conventional clip-wise approaches, which use video clips as input, require capturing over one hundred frames before inference, thus introducing several seconds of delay and hindering real-time use. Meanwhile, frame-wise approaches struggle to capture long-range temporal and periodic features of physiological rhythms, and therefore lead to reduced estimation accuracy. To overcome these issues, we propose StreamPPG, a unified architecture that enables low-latency frame-wise physiological signal estimation while achieving competitive accuracy compared with clip-wise approaches. StreamPPG is trained under a consistent privileged learning (CPL) strategy, which leverages ground-truth rPPG signals as privileged information to enhance the model's representation capability. Extensive experiments demonstrate that StreamPPG achieves state-of-the-art accuracy across multiple datasets while maintaining real-time throughput on edge devices.
