头部旋转运动下自适应非接触鲁棒性心率检测方法

基于人脸视频的生理信号检测面临的主要挑战是运动伪影噪声.针对受试者头部刚性旋转运动引起的伪影噪声,本文提出利用头部运动信息构建自适应滤波器的非接触式心率检测方法.该方法利用人脸二维和三维的特征点计算受试者运动中头部的偏航和俯仰欧拉角度,并将其作为调控过程噪声协方差的信号质量指数,进而构建了自适应Kalman滤波器,实现了稳健的心率估计.实验结果表明:本文提出的方法可有效抑制头部刚性旋转运动引起的噪声,平均绝对误差为2.22 beat/min,均方根误差为2.76 beat/min,与现有方法相比准确度分别提升9%与24.6%,具有统计显著性.本文提出的头部旋转角度自适应非接触鲁棒性心率检测方法在自发运动的真实场景下能有效提升检测的准确性,扩大了成像式光电容积描记技术在视频健康监测领域的使用场景.

Adaptive non-contact robust heart rate detection method under head rotation motion

The dominant challenge of vital signal monitoring based on facial video is to eliminate the interference of motion artifacts.In this paper,we propose a non-contact heart rate detection method based on an adaptive filter constructed by head movement information to tackle the noise of motion artifacts caused by the rigid rotation of the subject’s head.The two-dimensional and three-dimensional feature points of the subject’s face are used to calculate the yaw and pitch Euler angles of the head movement,then the yaw and pitch Euler angles are used as a novel signal quality index(SQI)for modulating process noise covariance to construct an adaptive Kalman filter,and finally robust heart rate is estimated by this method.The experimental results show that the proposed method can effectively suppress the noise caused by the head rigid rotation with an average absolute error of 2.22 beat/min and a root mean square error of 2.76 beat/min,which are statistically significant with an accuracy improvement of 9%and 24.6%,respectively,compared with the existing methods.The adaptive non-contact robust heart rate detection technique based on head rigid rotation may effectively enhance the accuracy in real-world motion situations,as well as broaden the range of applications for IPPG in the field of the video-based monitoring of health conditions.