基于相关系数与自适应滤波的心率估计

Heart Rate Estimation Based on Correlation Coefficient and Adaptive Filtering

近年来,基于光电容积脉搏波(PPG)实时监测心率受到各界的广泛关注,但由于在运动状态下光电容积脉搏波信号容易受到运动伪迹(MA)的叠加影响,尤其当运动幅度越大时,脉搏波信号产生的叠加干扰越严重,使得难以利用PPG信号准确计算出计心率值。为了保证能够从脉搏波信号中准确地估计心率值,为此提出一种去除脉搏波信号中运动伪迹的新算法CC-RLS,此算法利用复合归一化含噪脉搏波信号和均方根形式的加速度信号之间的相关系数法(CC)初步减少运动伪迹对PPG信号的叠加;选取合适的信号作为递归最小二乘(RLS)自适应滤波的输入信号再次滤除相关运动伪迹,增加了与心率相关的谱峰峰值,最终得到较为纯净的脉搏波信号,并且使用谱峰追踪算法进行实时心率估计。根据实验结果表明,相关系数法能够快速的消除运动伪迹谱峰峰值,并且在不同的运动状态下,该算法在10个PPG数据集上的心率估计误差较小、相比其他算法具有更少计算时间、较高的估计精度、算法稳定性高、鲁棒性强,并带来较强的实用价值。

In recent years, real-time monitoring of heart rate based on photocapacitance pulse wave (PPG) has received widespread attention from various fields. However, due to the fact that the photocapaci-tance pulse wave signal is easily affected by the superposition of motion artifacts (MA) during exer-cise, especially when the amplitude of exercise is larger, the superposition interference generated by the pulse wave signal becomes more severe, making it difficult to accurately calculate the heart rate value using PPG signals. In order to ensure accurate estimation of heart rate values from pulse wave signals, a new algorithm CC-RLS is proposed to remove motion artifacts from pulse wave sig-nals. This algorithm utilizes the correlation coefficient method (CC) between composite normalized noisy pulse wave signals and root mean square acceleration signals to preliminarily reduce the su-perposition of motion artifacts on PPG signals;Selecting a suitable signal as the input signal for re-cursive least squares (RLS) adaptive filtering, the relevant motion artifacts are filtered again, in-creasing the spectral peak value related to heart rate, and finally obtaining a relatively pure pulse wave signal. The spectral peak tracking algorithm is used for real-time heart rate estimation. Ac-cording to the experimental results, the correlation coefficient method can quickly eliminate the peak values of motion artifact spectra. Moreover, under different motion states, the algorithm has smaller heart rate estimation errors on 10 PPG datasets, has less computational time, higher esti-mation accuracy, high algorithm stability, strong robustness, and brings strong practical value compared to other algorithms.