Study On Non-contact Heart Rate Estimation Method Based On Facial Video

Heart rate can measure physiological and emotional state of human body,the traditional heart rate measurement usually uses high-precision contact equipment,such as pulse oximeter or electrocardiogram.However,contact heart rate measurement will bring many inconvenience.In recent years,researchers have proposed a variety of non-contact heart rate estimation methods based on face video,but the accuracy is not high.This article mainly studies how to extract more accurate pulse wave signals.Firstly,the heart rate representation ability of two common data sets,PURE and COHFACE,in different color space is studied.Then,the 3D convolutional network with residual structure is used to restore pulse wave signal of face video.The main content of this paper is as follows:1)We use Open Face to detect facial key points,connect the key points in specific locations below the eyes and above the lips,extract the region of interest frame by frame,which can reduce the background noise in the extraction process of pulse source signals.We study heart rate representation ability in different color space,with RGB,HSV,YCr Cb,Lab and YIQ color space,and using Eulerian Video Magnification algorithm to amplify the characteristic signal,finally it is concluded that HSV space effect best,the Mean Absolute Error(MAE)and the Root Mean Square Error(RMSE)tested on the COHFACE dataset reach 4.00 and 4.57 respectively.Therefore,we use this color space as the input of the subsequent three-dimensional convolution model2)The method of overlapping sampling and horizontal flipping is adopted to enhance the dataset,which can make the parameters of the three-dimensional convolutional network adjusted in the right direction,and finally the model can be better generalized.3)This paper proposes a non-contact heart rate estimation process based on three-dimensional convolutional network.In order to extract more accurate pulse wave signals,this paper designs three-dimensional convolutional pulse wave signal extraction algorithm,the Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)of this algorithm tested on the PURE dataset are 1.18 and 1.14 respectively,tested on COHFACE dataset are 3.94 and 3.65 respectively.4)This paper realizes non-contact heart rate measurement based on human face video.We test it in real scenes with 10 subjects,and the result shows that the error is less than 5%,which proves that the heart rate extraction method based on three-dimensional convolution in this paper can be applied in actual scene.Our paper’s method has certain feasibility,which can use deep learning methods for non-contact heart rate estimation in real world scenarios.