Enhancing Heart Rate Prediction Through Deep CNNs Using Facial Features From Non-Contact Video Analysis

With the increasing prevalence of chronic health conditions and the growing trend of telemedicine, there is a rising demand for reliable and non-invasive remote health monitoring solutions. The proposed paper aims to develop a real-time health monitoring system by leveraging advanced signal processing techniques and computer vision through webcam integration. The system focuses on the extraction and analysis of physiological signals, particularly those derived from facial features using CNN, to monitor and assess health parameters such as heart rate. The system employs a robust signal processing pipeline that includes color extraction, normalization, detrending, interpolation, and Fourier transformation (FFT) to analyze the periodicity of signals captured from facial regions of interest (ROIs). These signals, primarily focusing on the green color channel, are indicative of blood flow and can be used to estimate heart rate. Additionally, a Butterworth bandpass filter is applied to refine the signal, ensuring that only the relevant frequency components are retained for accurate analysis. The core of the project is a computer vision system that captures real-time video input from a webcam, processes each frame to extract the necessary facial regions using a CNN, and applies the aforementioned signal processing techniques to monitor physiological health indicators. The system is designed to function autonomously, requiring minimal user intervention, and provides real-time feedback on the user's health status. By integrating signal processing with computer vision, this project aims to create an accessible and non-invasive tool for continuous health monitoring, which can be extended to applications in remote healthcare, fitness tracking, and wellness monitoring.