基于FaceBoxes和ResNet34的人脸视频心率测量

Facial Video Heart Rate Detection Based on FaceBoxes and ResNet34

基于人脸视频的非接触式心率检测存在运动伪影和噪声干扰等问题,为克服运动伪影对心率检测的影响,文中提出一种基于FaceBoxes和改进ResNet34的人脸视频心率检测方法。对人脸视频帧进行人脸检测和特征点检测,能够准确定位每一帧的ROI区域,克服微小运动影响,提取ROI区域内RGB三通道信号,进行空间平均预处理、信号降噪,获得脉搏波信号,计算出心率。实验结果表明,基于改进ResNet34的特征点检测在人脸视频心率检测中发挥了良好的性能,在一定程度上克服了运动伪影的影响,并且提高了原有人脸视频心率检测方法的推理速度。

The non-contact heart rate detection based on facial videos has problems such as motion artifacts and noise interference.To overcome the impact of motion artifacts on heart rate detection,a facial video heart rate detection method based on FaceBoxes and improved ResNet34 is proposed in this paper.Performing face detection and feature point detection on facial video frames can accurately locate the ROI region of each frame,overcome the influence of small movements,extract RGB three channel signals within the ROI region,perform spatial averaging preprocessing and signal denoising,obtain pulse wave signals,and calculate heart rate.The experimental results show that the feature point detection based on improved ResNet34 performs well in facial video heart rate detection,overcomes the influence of motion artifacts to a certain extent,and improves the inference speed of the original facial video heart rate detection method.