Research Of Heart Rate Measurement From Near-Infrared Videos Based On Joint Blind Source Separation With Delay-Coordinate Transformation

Heart rate(HR),as one of the important vital signs of human body,can directly reflect the physiological health of the human body and mood fluctuations.HR is also an important indicator for the diagnosis of cardiovascular diseases.Accurate and effective HR monitoring is not only conducive to the early prevention of related diseases,reduces the incidence of disease,but also provides a strong basis for later treatment,which is of great significance to people's health.Traditional heart rate monitoring instruments require direct contact with human skin,which may cause discomfort or allergic reactions.It limits the application scope of heart rate monitoring.In recent years,non-contact HR measurement technology based on imaging photoplethysmography(i PPG)has become a research hotspot due to its features of low cost,easy to use,convenient,comfort,non-invasiveness,and wide application range.Up to now,most of the studies of contactless HR measurement based on i PPG technology have used RGB cameras.They works well with a stable external light source,whereas the red-green-blue(RGB)camera does not work well under the situations that the ambient light varies dramatically or is fully dark.In order to broaden the application fields,the near-infrared(NIR)camera is employed to HR measurement for illumination-dynamically-changing or dark situations.In this paper,a novel framework based on joint-blind-source-separation with delay-coordiante transformation,termed as JBSS-DCT,is proposed to evaluate HR from NIR videos.The main procedures are as follows:Firstly,face detection and tracking technology is used to identify multiple facial sub-regions of interest from the captured NIR video frames,and a single channel signal composed of pixel mean within each sub-region of interest is generated frame by frame.Secondly,the single channel signal is converted into multi-channel signals by delay coordinate transformation(DCT)and treated as a separate data set.The state space of the generated data set is equal to that of the original single channel signal that contains all dynamic variables.Thereby,several data setscorresponding to the multiple facial sub-regions of interest are generated.Then,independent vector analysis(IVA),one of the joint blind source separation(JBSS)methods,is used to extract the underlying source component vector(SCV)shared in multiple data sets,and the first ordering of the SCV containing the most correlated HR information are considered as the HR candidates.Finally,the power spectrum distribution(PSD)of each HR candidates was calculated,and the one with highest signal to noise ratio(SNR)was selected as the target HR source,the dominant of frequency of which was determined as that of the HR.Thereby,the HR was evaluated in the form of beat per minute(bpm).In this paper,the public database DROZY was used to verify the proposed method,and four other typical i PPG methods were employed for comparison.Experimental results demonstrated that the proposed method has more advantages.