Adaptive Parameter Optimization for Robust Remote Photoplethysmography

TL;DR

This paper introduces PRISM, a training-free method that adaptively optimizes rPPG signal extraction by jointly selecting color-mixing and detrending parameters online. By extending the POS framework with an online grid search over α and λ guided by a signal-quality objective, PRISM achieves state-of-the-art unsupervised performance on PURE and competitive results compared with supervised methods on UBFC-rPPG, all in real time without training. The approach is validated on two datasets with thorough ablations and toolbox enhancements, demonstrating robust HR estimation across varying lighting and motion. The work highlights the viability of online parameter adaptation to boost rPPG robustness and sets the stage for broader, deployment-ready unsupervised solutions.

Abstract

Remote photoplethysmography (rPPG) enables contactless vital sign monitoring using standard RGB cameras. However, existing methods rely on fixed parameters optimized for particular lighting conditions and camera setups, limiting adaptability to diverse deployment environments. This paper introduces the Projection-based Robust Signal Mixing (PRISM) algorithm, a training-free method that jointly optimizes photometric detrending and color mixing through online parameter adaptation based on signal quality assessment. PRISM achieves state-of-the-art performance among unsupervised methods, with MAE of 0.77 bpm on PURE and 0.66 bpm on UBFC-rPPG, and accuracy of 97.3\% and 97.5\% respectively at a 5 bpm threshold. Statistical analysis confirms PRISM performs equivalently to leading supervised methods ($p > 0.2$), while maintaining real-time CPU performance without training. This validates that adaptive time series optimization significantly improves rPPG across diverse conditions.

Figures (1)

• Figure 1: Radar plots comparing performance across five evaluation metrics on the PURE and UBFC-rPPG datasets. PRISM outperforms other methods across most of the metrics. All metrics are normalized such that larger area indicates better performance.