Research On Non-contact Heart Rate Measurement Method Based On Facial Video

Heart rate is one of the most important indicators of the human health and has been widely used in the diagnosis of diseases.Previous heart rate measurement methods are based on traditional contact measurement methods,such as ECG monitoring.However,these contact measurement methods need to run on a large number of complex hardware and usually require direct contact with the human body,which may cause discomfort for a long time.Today,affected by the COVID-19 pandemic,contact measurement methods may cause a large number of infections.so non-contact heart rate measurement methods have become a research hotspot,especially heart rate measurement methods based on remote photoplethysmography(r PPG).The r PPG method can directly measure heart rate through facial video without complicated equipment,which has significant advantages.However,the r PPG method has problems such as being easily affected by the background and insufficient extraction of spatial and temporal features.In response to these problems,this paper conducts an in-depth study on the r PPG heart rate measurement method.The main research contents are as follows:(1)Aiming at the problems that r PPG signals are easily affected by the background and are sensitive to temporal information,a network model of r PPG heart rate measurement based on temporal central difference convolution is proposed,the proposed model is mainly composed of stacked temporal central difference convolution.Since the r PPG signal is directly reflected in the change of facial skin pixels over time,the traditional 3D convolution cannot capture the time difference information well,while temporal central difference convolution considers the temporal gradient cues,which can better capture temporal context information for recovering r PPG signals.The input of the model is a sequence of high-quality region of interest(ROI)spliced by cheek and forehead,which are defined by face keypoint detection.The cheek and the forehead are proved to be the two areas with rich r PPG information and the largest absolute size of the human face.After this processing,the influence of non-skin pixels on the face and background on heart rate measurement can be minimized.The experimental results show that the mean absolute error of the method proposed in this chapter is smaller than that of the same type of algorithm in heart rate measurement,which can effectively improve the accuracy of heart rate measurement.(2)Aiming at the problem of poor spatiotemporal feature extraction and long-term context learning ability of the above model,on the basis of this model,a network model of r PPG heart rate measurement based on multi-scale fusion and attention mechanism is proposed.Aiming at the problem of insufficient spatiotemporal feature extraction,the model uses a multi-scale module fused with an attention mechanism,extracts features by using different time scale convolution kernels,then fuse them,and finally use the attention mechanism to characterize them according to the importance of channels and spatial positions.Aiming at the problem that the network model has poor ability to learn long-term context,the temporal relationship enhancement module proposed in this paper uses convolutional long short-term memory network to better learn the relationship between frames to enhance the network’s long-term learning ability.The experimental results show that the improved model proposed in this chapter achieves better results,which can further reduce the error.(3)A real-time video streaming r PPG heart rate measurement system is designed and implemented.The system can directly collect facial video through the camera,and input it into the model proposed in this paper to predict the r PPG signal after preprocessing ROI,and finally calculate the heart rate and display it dynamically in real time.The system can visually display the selected region of interest,predicted r PPG signal and heart rate.The experimental results of the system verify the feasibility of the method proposed in this paper for real-time heart rate measurement.