What question did this study set out to answer?

The research aims to evaluate machine learning techniques for quality classification of remote photoplethysmography signals in vehicles.

February 23, 2026Open Access

View Full Paper

Machine learning-based remote PPG signal quality assessment for in-vehicle driver monitoring

SBSonja BabacPhilips (Finland)LVLuc P.J. VostersPhilips (Finland)RVRik VullingsElectrophysiology

Key Points

The research aims to evaluate machine learning techniques for quality classification of remote photoplethysmography signals in vehicles.
Systematic evaluation of machine learning and deep learning techniques for rPPG quality classification.
Involves machine learning methods like XGBoost, support vector machines, and deep learning with convolutional neural networks.
Real-world validation conducted with 31 drivers across various driving scenarios.
XGBoost using five frequency-domain features achieved an AUC of 0.86.
Peak-centered magnitude spectrum significantly improved the classification ability compared to traditional time-domain approaches.
The approach demonstrated low computational cost measured in floating-point operations per second.

Abstract

• First systematic evaluation of ML and DL for in-vehicle rPPG quality classification. • Peak-centered magnitude spectrum enhances rPPG quality classification. • XGBoost with five frequency–domain features achieves AUC 0.86 while keeping computational efficiency low. • Real-world validation on 31 drivers under varied driving scenarios. Camera-based vital signs monitoring is emerging as a key component of intelligent transportation systems, enabling unobtrusive driver health tracking to enhance road safety. Remote photoplethysmography (rPPG), a technique enabling this camera-based monitoring, allows pulse rate estimation; however, its use in vehicles is susceptible to motion artifacts and dynamic lighting conditions, limiting its reliability. To address this, we evaluated various machine learning (ML) and deep learning (DL) techniques for rPPG signal quality classification, including extreme gradient boosting (XGBoost), support vector machines, convolutional neural networks, and long short-term memory. Through a rigorous comparison of time and frequency domain representations, we reveal that the peak-centered magnitude spectrum markedly enhances the ability to distinguish reliable from unreliable rPPG segments, surpassing traditional time-domain approaches. Notably, our XGBoost model, leveraging the top five frequency-domain features, not only achieved superior performance (AUC of 0.86, with strong sensitivity) compared to existing state-of-the-art PPG quality classification methods on our dataset, but also demonstrated vastly lower computational cost quantified by floating-point operations per second. To our knowledge, this is the first study to systematically investigate ML and DL approaches for rPPG signal quality classification, particularly within a real-world automotive context. These findings pave the way for integrating robust, computationally efficient driver health monitoring into advanced Driver Monitoring Systems and next-generation remote health monitoring technologies.

AI से पूछें

Bookmark

View Full Paper

Cite This Study

Babac et al. (Sat,) studied this question.

synapsesocial.com/papers/699ba0a772792ae9fd8708cd https://doi.org/https://doi.org/10.1016/j.bspc.2026.109869

AI से पूछें

Bookmark

View Full Paper