The Research And Implementation Of Non-contact Blood Volume Pulse Detection System Based On Facial Videos

With the development of PPG technology,more and more researchers carried out research on non-contact physiological signal detection.We can separate blood volume pulse(BVP)and eyes movement signal from facial videos by using different signal processing methods,and then realize the telemetry of heart rate,respiratory rate,blood oxygen saturation,heart rate variability and blood pressure and other vital signs parameters,and these parameters can accurately reflect the human body’s health level and emotional state.The non-contact physiological signal detection method is of important research value in the field of health care,fatigue test and emotion recognition.At present,there are two kinds of methods of the non-contact physiological signal detection based on PPG technology.One is only using the green channel data of facial video to extract the BVP called G-BVP,another is using ICA to separate BVP from the R,G,B signals called ICA-BVP.The BVP signals from two kinds of methods are different,these differences have an effect on subsequent processing steps.However,existing research dose not demonstrate the advantage and the weakness of the two methods.This thesis compared the performance and characteristics of ICA-BVP with G-BVP in different scenarios.The main work of this thesis is summarized as follows:1)Firstly,this thesis compares face detection algorithm with Mean Shift algorithm in the pretreatment process.It analyzes the original signal and BVP signal extracted from facial video and then chooses the Mean Shift algorithm for facial tracking.Then several kinds of ICA algorithms are studied,and the effects in BVP extraction and vital signs parameters calculation are compared.Considering the execution efficiency and platform portability of these ICA algorithms,SOBI algorithm is chose for ICA-BVP method in the experiment.2)The performance and characteristics of ICA-BVP and G-BVP in different scenarios are compared after learning the existent interferences caused by regional choice,blinking and illumination change based on facial videos.3)In the specific implementation process of the ICA-BVP method,a spectral kurtosis based automatic identification of BVP signal is proposed to solve the inherent shortage of ICA output ambiguity,which is effective in dynamic BVP extraction and vital signs parameters calculation.4)A non-contact BVP extraction system based on Visual Studio 2013 is realized.It can be used in dynamic heart rate estimation online through a webcam or offline through a video file.Users can choose G-BVP or ICA-BVP to extract BVP signal according to the situation in order to analyze the health condition further.