Enhancing Robustness of Remote Heart Rate Estimation In Varied Illumination Environments

This research enhances ML models for remote heart rate estimation under varied illumination using an extensive rPPG dataset. We employ two approaches: feature analysis with CSV data and deep learning-based image analysis. For CSV data, advanced ML algorithms (Random Forest, Gradient Boosting, SVR) are used, comparing classical methods like ICA with our Feature Correlation Model. For image data, a custom CNN is developed and benchmarked against traditional rPPG methods like CHROM and Green. This evaluation highlights the effectiveness of these techniques in detecting subtle skin color changes for heart rate monitoring.Our proposed architecture combines insights from CSV and image data analyses to create a robust framework for non-contact heart rate monitoring. It uses a dual-model approach: the Feature Correlation Model leverages CSV data to highlight physiological signal interactions, while the CNN processes image data to capture spatial-temporal features. This integration enhances predictive capabilities and adaptability across diverse conditions, advancing rPPG technology in healthcare, telemedicine, fitness tracking, and human-computer interaction. The goal is to improve prediction accuracy and reliability, addressing the complex dynamics between physiological features and varying illumination, and setting a new standard for remote photoplethysmography applications.