Contact-Free Heart Rate Measurement From Human Face Videos and its Biometric Recognition Application

Authors Organisations
Type

Student thesis: Doctoral ThesisDoctor of Philosophy

Original languageEnglish
Awarding Institution
Supervisors/Advisors
Award date2019
Links
Show download statistics
View graph of relations

Abstract

The heart is the most important muscular organ of a human body and its strength can be assessed by the rate at which it beats [1]. Heart rate measurement plays an important role in human health assessment and there have been a number of methods suggested in order to monitor it remotely with more ease and comfort. Contact-free heart rate measurement is one way to make it user-friendly and can also be used for covert surveillance. The performance of previously developed touch-free methods was found to be efficient and accurate under controlled conditions. In realistic and more challenging scenarios, their performance is degraded, as each method has its
limitations. Typically, the performance is dependent on controlled lighting and limited subject movement. More realistic situations require more robust contact-free ways to measure the heart rate. Our work aims to obtain a good understanding of the underlying problems of non-contact pulse estimation. We have proposed a method that can overcome many of the problems of light reflection and subject’s movement while measuring the heart rate from human faces remotely using an ordinary webcam. We use Saragih face tracker [2] to track the faces from the recorded videos which provides more reliable extraction of a region of interest (ROI) than the simple face detection. We find the robust mean of the skin pixel’s color values of the selected ROI which can improve the measurement of the skin color in each frame, reducing the impact of facial hair, wrinkles, and lighting. In addition, we calculate the least-squares error optimal filter using our training dataset to estimate the heart rate more accurately from
the measured color changes over time. These methods not only improved the accuracy of heart rate measurement but also resulted in extraction of a cleaner pulse signal, which can be integrated into many other useful applications. We explore whether this technique has the potential for advancing human biometric systems employing the pulse signals obtained. We have presented a method for the biometric recognition application of the pulse signals obtained from the facial videos. The Radon transform images of the pulse signals obtained from facial imagery were used for the extraction of distinctive features, and the decision tree and ANNs machine learning techniques were employed to classify those features for identification purposes. Results obtained show that our proposed contact-free heart rate measurement method has significantly
improved on existing methods, and that biometric recognition systems can be improved with the use of the techniques we propose