Research on nighttime IPPG algorithm based on ROI delay expansion and fundamental frequency constrained FastICA.

This study aims to enhance the accuracy and reliability of imaging photoplethysmography (IPPG) for heart rate (HR) measurements during nighttime by introducing an innovative approach that combines fast independent component analysis (FastICA) with aime-elayedulti-imensionalxtendedegionsfnteresttraction () technique, specifically tailored to overcome the challenges posed by motion artefacts and the difficulty in identifying regions of interest (ROIs).This research employs a dual-method strategy for the precise extraction of ROIs and robust processing of HR signals in nighttime IPPG scenarios. Initially, a face detection algorithm is integrated with a grayscale clustering technique to pinpoint optimal ROIs. This is followed by the application of the mutual information delay method to synthesize multi-channel IPPG signals. Concurrently, the'sundamentalrequency is leveraged as a prioronstraint within the iterative process of(), mitigating the susceptibility to initial value fluctuations inherent in FastICA. The synergistic application of these methodologies substantially bolsters the stability and robustness of nighttime HR measurements, particularly in conditions characterized by significant motion.The efficacy of the proposed method, which incorporates HRFFC-FastICA, is initially validated through performance testing using the MR-NIRP dataset. This step serves to assess the practicality of the approach for nighttime IPPG HR measurements. Following this, a series of modular ablation studies and comparative evaluations against current nighttime IPPG algorithms are executed. The statistical outcomes demonstrate that our method achieves a mean absolute error (MAE) of 4.57 beats per minute (bpm) and a root mean squared error (RMSE) of 5.95 bpm. In direct comparison with prominent algorithms such as SparsePPG and PhysNet, the method exhibits a notable enhancement in MAE by up to 8.39 bpm and a significant decrease in RMSE by 17.83 bpm. The 95% confidence interval of the Bland-Altman graph of this method is between 9.5 and -12.8 bpm. Compared to other comparable methods, this interval is significantly narrower, with a width nearly half that of alternative approaches, indicating superior precision and reliability.The significance of this research is highlighted by the experimental outcomes that demonstrate the considerable advantages of the TDMDE-ROI-Ex method. This technique significantly reduces reliance on facial motion, which is crucial for accurately identifying facial skin colour regions of interest. Moreover, integrating the HRFFC-FastICA method effectively counteracts the effects of motion artefacts and the initial value sensitivity inherent in the FastICA process. The introduction of this methodology into nighttime IPPG monitoring significantly strengthens the system's robustness and stability, thereby extending the range of IPPG technology applications and improving its overall measurement performance.