rFaceNet: An Advanced Method for Enhanced Physiological Signal Extraction Through Facial Contours
Core Concepts
rFaceNet enhances facial BVP signal extraction by integrating identity-specific facial contours, improving interpretability and performance.
Abstract
Abstract:
rFaceNet introduces an advanced method for extracting facial BVP signals with a focus on facial contours.
It efficiently extracts facial contours using a Temporal Compressor Unit (TCU) and Cross-Task Feature Combiner (CTFC).
Introduction:
Traditional heart rate extraction methods are being replaced by non-contact photoplethysmographic techniques like rPPG.
Previous solutions failed to integrate facial contours due to technical limitations.
Methodology:
rFaceNet framework includes TCU for temporal compression and CTFC for feature combination.
Experiments:
Intra-dataset and cross-dataset heart rate estimation experiments demonstrate the effectiveness of rFaceNet.
Ablation Study:
Combining TCU and CTFC components in rFaceNet yields optimal results.
rFaceNet
Stats
Remote photoplethysmography (rPPG) extracts blood volume pulse (BVP) signals from video frames.
Extensive experiments show superior performance of rFaceNet in heart rate estimation benchmarks.