Research Of Non-contact Heart Rate Detection Based On Facial Videos

Heart rate(HR)is an important physiological indicator that reflects the health status of human body.In the medical and health field,HR parameter plays a great role in the prevention and treatment of cardiovascular diseases.Therefore,being able to achieve accurate daily HR detection is crucial for human health assessment and monitoring.Most of the HR detection methods nowadays mainly require professional equipment to directly contact with the skin of the subjects.At the same time,the long-term contact may also give rise to the discomfort of patients in practice.And it is also difficult to meet the needs of daily HR monitoring.Recent research shows that the image Photoplethysmography(iPPG)has the potential of non-contact physiological signal detection.We can use different signal processing methods to extract from the facial video the biological signal of the Blood Volume Pulse(BVP)that reflects the change of the blood volume in vessels.The HR parameters can be further obtained based on BVP signal.Based on iPPG technology,several research institutions have carried out research on the extraction of physiological parameter such as HR.However,in the published relevant research literature,most of the facial video data were collected when the subjects are in an ideal state of relative stillness,that is,through the design of experimental paradigm and the motion artifact interference is minimized.When the subject's head moves,the accuracy of HR detection will decreases rapidly as well as the robustness of the algorithm has obvious flaw.Meanwhile,in the research of contactless heart rate detection based on facial video,there is no uniform standard for color space selection.Thus,this thesis carried out research on anti-motion interference and color space selection on the basis of facial video HR detection technology,and proposed relevant solutions as well as achieved relatively ideal experimental results.The main research work of this thesis is as follows:(1)Base on the analysis of a large number of experimental data and related research literature,the qualitative and quantitative evaluation index of the HR detection algorithm performance were given.Meanwhile,we use this index to evaluate the validity of the antimotion inference method presented in this thesis and compare the consistency of results by HR reference values and our new method.(2)Aimed at the interference in the HR detection,this thesis puts forward a new BVP detection method based on the Kanade-Lucas-Tomasi(KLT)and Independent Component Analysis(ICA).We use the eye detection and KLT to locate and track the Region of Interest(ROI)and ICA to improve the signal-to-noise ratio.In order to verify the effectiveness of the algorithm,this thesis compares it with other methods in the analyses of 120 facial videos.The experimental result proves that this method outperforms others in the elimination of movement interference.In the manipulation of Blind Source Separation(BBS),this thesis compares four common ICA methods and then presents the conclusion that Second Order Blind Identification(SOBI)is the best algorithm in extracting BVP signal.(3)In order to study the influence of color space selection on the accuracy of HR detection.Based on the HR detection algorithm proposed in this thesis,the characteristics of five commonly used color Spaces in heart rate detection accuracy were compared.And the public database LGI PPGI Face Video were used to conduct the HR detection experiment in different color spaces.The result indicates that the color space RGB performs well in the HR detection precision and computation complexity,etc.(4)On the basis of the anti-motion inference HR detection algorithm,we use C++ programming language and OpenCV to realize the offline and online anti-motion inference HR detection system based on Visual Studio 2013 platform.The HR detection test was carried out based on the software platform,and the test results showed that the results of the system's HR detection and the HR reference value had a good consistency.Finally,the operation interface of HR detection system is demonstrated.