ABSTRACT
This article aims to predict vital signs like heart rate (HR), respiration rate, and arterial oxygen saturation using ambient light video, eliminating chronic distortions through improved frame quality with BER estimation. The study employs the cascade residual CNN-FPNR technique for preprocessing and SNR enhancement using energy variance maximization. The image cascade network (ICNet) facilitates segmentation, achieving strong segmentation in low-light ambient videos. Remote photoplethysmography (iPPG) enables noncontact vital sign monitoring, predicting HR and respiratory rate (RR). An innovative noninvasive temperature and cyclical algorithm, incorporating principal component analysis and fast Fourier transform, evaluate patient HR and RR. To address challenges related to involuntary movements, a dynamic time-warping-based optimization method is used for precise region selection. The study introduces an intensity variance-based threshold analysis for arterial oxygen saturation level determination. Ultimately, the support vector machine (SVM) classification technique evaluates the ground truth, showcasing the system's promising potential for remote and accurate vital sign assessment.