PREHEAT: Precision heart rate monitoring from intense motion artifact corrupted PPG signals using constrained RLS and wavelets

Abstract

Abstract Goal Monitoring heart rate (HR) from intense motion artifact (MA) corrupted photoplethysmography (PPG) signals is a challenging problem. Accurate HR estimation from intense MA corrupted PPG signals can be very useful in HR monitoring programs during physical exercises. Method In this paper, we present a constrained recursive least-squares (cRLS) based denoising technique to effectively eliminate MAs from a composite PPG signal, and a coupled wavelet-Fourier based frequency estimation technique to estimate the HR more precisely. To separate the clean PPG, the MA corrupted composite PPG signal is decomposed using the ensemble empirical mode decomposition (EEMD) based signal separation technique. When EEMD fails to effectively separate the clean PPG from MAs, the new cRLS based denoising is used that can eliminate MAs even when there is an acceleration peak close to the HR peak. In HR estimation from the denoised signal, a novel wavelet based frequency estimation technique is incorporated along with the conventional Fourier based technique that improves the precision of HR tracking, especially from nonstationary PPG window frames. The autoregressive (AR) model based predictor and some heuristic decision based approaches are included for further reduction of the HR estimation error. Result The method is tested on 12 different subjects running on a treadmill with a peak velocity upto 15 km/h and we have obtained a mean absolute error of 0.81–0.84 beat per minute (BPM) with a standard deviation of ±0.82–0.96 BPM. Conclusion The proposed technique has achieved a benchmark result on these subjects which is much better as compared to the recently reported techniques in the literature. Significance The proposed method is expected to find a good application in wearable gadgets to monitor HRs accurately during fitness programs and other activities with intense physical motion.