Research On Key Technology Of Human Physiological Signs Extraction Based On Imaging Photoelectricplethysmography

The measurement for physiological parameters is an important means in clinical vital signs monitoring.With the rapid growth of advanced sensors,the technologies for measuring physiological parameters have begun to move forward from traditional contact measurement to non-contact method,which turns the attention from the equipment to the "person",that is,the subjects could carry out the measurement without being bound by special equipment.Imaging photoplethysmography(IPPG),a camera-based technique developed to measure physiological parameters in non-contact way rising in recent years,has gradually gained research interest.This technique collects video data of the naked skin surface and then reconstructs the physiological parameters from the image frames by relying on intelligent information processing algorithms.Different from traditional measurement methods,the IPPG-based measurement was limited by several objective defects,such as the complex background noise in observed signals and poor robustness of the algorithms for physiological parameter extraction.These defects resulted in slow progress of relevant researches.On basis of in-depth study of the IPPG-based methods for physiological parameters extraction,this thesis improved the key technologies involved in this field by the theoretical analysis and related algorithm designs.Through comprehensive comparative evaluation on blood volume pulse(BVP)extraction by blind source separation(BSS)/ independent component analysis(ICA)approaches,the lack of this important work in existing researches was filled.On this basis,the IPPG-based measuring methods of several physiological parameters were explored and improved,such as blood oxygen saturation(Sp O2),respiratory rate(RR),heart rate variability(HRV)and other indexes.The main work and innovations of this thesis can be summarized into the following four parts:First of all,a comparative evaluation of BVP extractions by different BSS/ICA algorithms was performed in the thesis,to provide valuable references for optimal algorithms selection and the further exploration in IPPG.The main work included: 1)comparison and optimization of the computational costs;2)stability of BVP source permutation;3)Noise residue in BVP source signal;4)stability of BVP extraction by pre-trained spatial filter.A significant finding was obtained in the experiment that the second order blind identification(SOBI)was generally superior to other algorithms in quality of BVP source signals extracted from RGB videos.Aiming at the ambiguity of source sorting in BSS/ICA,the method based on kurtosis of power spectrum combined with HR predicted values was designed,to obtain accurate BVP source identification.Secondly,a SOBI-based method for Sp O2 extraction was proposed in this thesis.We carried out an in-depth examination of the traditional webcam-based Sp O2 extraction methods.On this basis,rather than roughly using the standard deviations(STD)of AC components of red and blue channels for Sp O2 calculations,we performed separation for the AC components by using SOBI,and then used the energy coefficients retained in the mixed matrix to replace the STD required in the algorithm,to avoid the problem of complex noise faced by the traditional methods.Moreover,the steady reference data was introduced to compensate for the defect that the change trend of Sp O2 concentration was easy to lose in using red and blue channel combination.Through these efforts,the anti-noise capability of the algorithm was significantly enhanced,and the related defect was compensated for.The experimental results indicated that the proposed method produced reliable Sp O2 estimation that could potentially—with further research—be used in real applications.Thirdly,we proposed the methods for extracting respiratory components and other features based on facial motion artifacts.The rhythmic components contained in motion artifacts that appeared in some facial regions were extracted as important feature parameters.On this basis,we achieved the synchronous extraction of respiratory component and BVP from the dual region of interest(ROI)composed of mouth and throat regions by using SOBI separation.Moreover,on the basis of motion sensitivity analysis of the RGB observed signals,the blinking pulses and yawn features were extracted from the R-channel.This work could expand the application of IPPG and provide a convenient and efficient information channel for many fields,such as human-computer interaction control.Fourthly,we explored the solution to extract HRV parameters from RGB signals depending on constrained ICA(c ICA)and variational modal decomposition(VMD)algorithms.The work mainly involved BVP reference signal construction,BVP source signal extraction based on c ICA,high-quality pulse component extraction,HRV parameters estimation and so on.In the study,the combination of c ICA and VMD realized the extraction of high-quality pulse wave components,which solved the problem that HRV parameters were hard to estimated due to the defect—low signal-to-noise ratio(SNR)of BVP components in traditional researches.Finally,through the theoretical analyses,algorithm designs and experimental verifications in the above work,the accuracy and reliability of target signs measurements were significantly improved.Further,the exploration for some key technologies in IPPG could provide valuable conclusions or ideals for this field,and expand its application prospects.