光学 精密工程, 2020, 28 (3): 548, 网络出版: 2020-05-12   

基于人脸视频的心率参数提取

Extraction of heart rate parameters from video of human face
作者单位
1 郑州大学 电气工程学院, 河南 郑州 450001
2 郑州大学 体育学院, 河南 郑州 450001
3 郑州大学 生命科学学院, 河南 郑州 450001
摘要
为了在舒适非接触环境下检测被试者的心率变化, 本文设计了一种通过普通摄像头来检测心率参数的信号处理系统。首先, 将KLT(Kanade-Lucas-Tomasi)算法跟踪识别到的人脸视频图像转换到YCbCr颜色空间来进行皮肤检测, 并同时转换到Cg颜色通道来提取高质量的光电容积脉搏波 (Photoplethysmography, PPG)信号。然后, 用Morlet复小波作为母波绘制PPG信号的小波能量谱图。最后根据心率信号的生理特性, 去除伪点噪声, 提取随时间变化的心率参数。实验结果表明, 该方法在静息状态下的测量结果同标准仪器测量结果的平均绝对值误差|Me|小于2 bpm(beats per minute), 误差的标准差SDe小于2.5 bpm, RMSE均小于2.6 bpm; 头部运动状态下两种测量方法的|Me|均小于2.3 bpm, SDe均小于2.9 bpm, RMSE均小于2.9 bpm。对两种测量方法进行Bland-Altman一致性分析, 其测量结果显示静息状态下差值的均数为0.295 7 bpm, 95%置信区间为-3.340 1~3.931 4 bpm; 头部运动状态下为0.383 2 bpm, 95%置信区间为-3.677 1~4.443 5 bpm, 表明本文提出的非接触式方法的测量结果同标准仪器的测量结果具有高度的一致性。
Abstract
To detect the heart rates of subjects in a comfortable non-contact environment, this study designs a signal processing system that can detect heart rate parameters using an ordinary camera. First, the face image captured by the Kanade-Lucas-Tomasi algorithm is converted into the YCbCr color space for skin detection. Simultaneously, the face image is converted to the Cg color channel to extract a high-quality photoplethysmography (PPG) signal. Then, Complex Morlet is used as the master wave to draw the wavelet energy spectrum of the PPG signal. Finally, according to the physiological characteristics of the heart rate signal, the pseudo-point noise is removed and the time-varying curve of heart rate parameters is extracted. Compared with the measurement results of the standard instrument, the mean absolute error (|Me|) of all the testers is less than 2 bpm (beats per minute), the standard deviation of error (SDe) is less than 2.5 bpm, and the root mean square error (RMSE) is less than 2.6 bpm in the resting state. The |Me| of all the testers is less than 2.3 bpm, while the SDe is less than 2.9 bpm, and the RMSE is less than 2.9 bpm in the head moving state. A Bland-Altman consistency analysis is performed for the two measurement methods. The results show that the mean of the difference () is 0.295 7 bpm and the 95% confidence interval is from -3.340 1 bpm to 3.931 4 bpm in the resting state; is 0.383 2 bpm and the 95% confidence interval is from -3.677 1 bpm to 4.443 5 bpm in the head moving state. Thus, it is confirmed that the measurement results of the non-contact method proposed in this paper are highly consistent with the measurement results of the standard instrument.

李晓媛, 武鹏, 刘允, 司红玉, 王振龙. 基于人脸视频的心率参数提取[J]. 光学 精密工程, 2020, 28(3): 548. LI Xiao-yuan, WU Peng, LIU Yun, SI Hong-yu, WANG Zhen-long. Extraction of heart rate parameters from video of human face[J]. Optics and Precision Engineering, 2020, 28(3): 548.

本文已被 3 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!