Improved peak detection technique for robust PPG-based heartrate monitoring system on smartphones

Abstract

Heartrate monitoring can be very decisive in terms of detecting heart-related diseases in early stages. Given that smartphones are used almost ubiquitously by humans on a daily basis and that they are equipped with several sensors and a powerful CPU, they are the ideal tools to accomplish daily heartrate monitoring. Many studies have shown the possibility of monitoring the heartrate in a very accurate way by using smartphone camera and the photo-plethysmography technique. However, in real cases, when using a smartphone camera as the input sensor, the pulse signal usually contains noise caused by fingertip movements, pressure changes between the smartphone camera and the fingertip, or changes in ambient light intensity. Hence, many techniques have been proposed to filter and detect real peaks in the pulse signal and avoid noise peaks. Most of those state-of-the-art techniques rely on the assumption that real peaks are those having the maximum values in their respective cycles. In some cases, the assumption is not correct, which decreases the accuracy of peak detection. Hence, this paper proposes a novel method to detect a pulse signal’s real peaks. The proposed method first smoothens a pulse signal by using a second-order Butterworth band-pass filter over a wide frequency range. Then, it considers the local minima of the signal as peaks and builds an optimization tree based on statistical properties, such as uniformity of peak time locations, peak count within a period, and variance in times elapsed between adjacent peaks, to detect combinations of peaks that optimize the heartrate computation. Results from experiments conducted using synthetic and real signals show that the proposed method can detect a pulse signal’s real peaks with higher precision as compared to conventional methods, and is very robust to signal noises. The proposed method has a significant heartrate estimation accuracy in various real scenarios and across smartphone models; its error rate is as low as 0.8588±0.0653 beats per minute, even in cases of signals with extensive noise.