Research On Non-contact Heart Rate Measurement Method Based On Deep Learning

Heart rate is one of the important indicators of human physiological and psychological activities,which can reflect the metabolic level of the human body,the health status of the nervous system and cardiovascular system.Traditional heart rate measurement requires the use of related equipment to make contact with the human body,and long-term wear can cause discomfort to the human body.In recent years,with the development of remote photoplethysmography(r PPG),video-based noncontact heart rate measurement methods have gradually become a popular research topic.This method can extract the heart rate signal from the face video without contact with the human body,and has the advantages of convenience,non-invasiveness,and low cost.This paper focuses on the existing problems of this method and studies a non-contact heart rate measurement method based on deep learning,aiming to improve the accuracy and stability of non-contact heart rate measurement.The main research content is as follows:(1)Aiming at the problem that some deep learning-based methods cannot consider time-domain features when use 2D convolution for calculation and the low robustness of the fixed Region Of Interest(ROI)for heart rate measurement,this paper proposes a non-contact heart rate measurement method based on 3D residual attention networks.This method uses 3D convolution for calculation and embeds a 3D Convolutional Block Attention Module(3D-CBAM)in its spatiotemporal convolution module,Strengthen the extraction of video channel features,time-domain features,and spatial features by neural networks,so that the model pays more attention to areas with strong physiological signal distribution.At the same time,introduce residual structures in this module to improve the performance of the model without changing the network depth.(2)Aiming at the problem that the model generalization ability is not strong due to the small amount of data in the existing datasets and the differences in imaging equipment,capture environment,human body state and skin color among the datasets,this paper proposes an improved non-contact heart rate measurement method based on spatio-temporal feature maps.Firstly,the traditional r PPG signal extraction algorithm is used to extract the corresponding initial r PPG signals from multiple ROI blocks to weaken the influence of data set differentiation,then the spatiotemporal feature map is constructed for the extracted signals,and finally the improved Res Net-18 feature fusion model is used to extract more accurate r PPG signal is extracted from the spatiotemporal feature map.(3)Using the improved heart rate measurement method based on spatio-temporal feature maps,this paper designs and implements a video-based non-contact heart rate measurement system based on Python,QT and Open CV,and measures different people to verify the effectiveness of the method proposed in this paper.(4)This paper has been tested on multiple public datasets and compared with some typical noncontact heart rate measurement methods,demonstrating the feasibility and superiority of the proposed method.