凌六一 汪晴晴 周孟然

摘 要：为了降低照明变化、相机内部等噪声干扰对远程光电容积描记法检测心率的误差，提出基于自适应噪声完备集合经验模态分解（CEEMDAN）与快速独立成分分析（FastICA）相结合的去噪方法，实现非接触式心率准确检测。首先对视频图像中人脸68个关键点进行检测，获得感兴趣区域（ROI）并得到源信号;然后通过CEEMDAN对源信号进行分解并选取合适频段内的固有模式函数（IMF）进行重构，对重构信号进行FastICA盲源分离后将RGB各通道信号进行快速傅里叶变换，最终利用频谱峰值对应频率计算得到心率。对10名不同性别和年龄的人员进行了实验测量，并将测量结果与脉搏血氧检测参考仪器进行了对比分析，测量均方根误差和平均绝对误差分别为0.72bpm 和0.60bpm。实验结果表明，所提方法能够有效去除心率检测源信号中的噪声，提高心率检测的准确性。

关键词：远程光电容积描记法;心率检测;CEEMDAN;FastICA

中图分类号：TP391 文献标志码：A

文章编号：1672-1098（2021）02-0001-08

收稿日期：2020-12-15

基金项目：安徽省科技重大专项计划项目（201903a07020013）

作者简介：凌六一（1980-），男，安徽枞阳人，教授，博士，研究方向：檢测技术与智能信息处理。

Non-contact Measurement of Heart Rate Based on CEEMDAN-FastICA

LING Liuyi1，2，WANG Qingqing1，ZHOU Mengran1

（1.School of Electrical and Information Engineering， Anhui University of Science and Technology， Huainan Anhui 232001， China;2.School of Artificial Intelligence， Anhui University of Science and Technology， Huainan Anhui232001， China）

Abstract：The accuracy of heart rate measurement based on remote photoplethysmography （rPPG） usually decreases due to the noise caused by lighting fluctuation and camera defects. In order to measure heart rate accurately， a de-noising method based on the complete ensemble empirical mode decomposition with adaptive noise （CEEMDAN） and fast independent component analysis （FastICA） is proposed， which is used firstly to obtain the region of interest （ROI） as well as source signal by establishing 68 key points in the face image from video stream， secondly to decompose the source signal with the CEEMDAN and then reconstruct it by making use of intrinsic mode function （IMF） within the appropriate frequency band， thirdly， to implement the based-FastICA blind source separation on the reconstructed signal to get independent RGB signal， and finally， to calculate the heart rate with the frequency， which is corresponding to magnitude peak in the frequency spectrum gotten by Fourier transform on the independent RGB signal. The heart rates of ten people of different gender and age are measured by the proposed method. The results are compared with those measured by standard pulse oximetry instrument， and the root mean square error （RMSE） and mean absolute error （MAE） are 0.72 bpm and 0.60 bpm respectively. The experimental results demonstrate the proposed method capable of effectively reducing the noise in the source signal and improving the accuracy of heart rate detection.

Key words：remote photoplethysmography; heart rate measurement; CEEMDAN; FastICA

人的心率（heart rate，HR），即心脏跳动的频率，是临床诊断和生命体征监测中最重要的参数之一。人类心率一般为45～120次/min，传统的接触式心率检测要求患者佩戴胸带或电极，远程光电容积描记法（remote photolethysmogracphic，rPPG）使用摄像机就可以监测心率[1]。然而，rPPG技术的最大问题是噪声干扰，测量过程中照明变化、相机内在噪声以及被测者的运动，包括头部运动、面部表情、眨眼和说话等，这些都会给心率检测带来一定误差。……