Dynamic heart rate measurements from video sequences using canonical component analysis / Ling Siong Shi

• Main Author: Ling, Siong Shi
• Format: Thesis
• Published: 2019
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://studentsrepo.um.edu.my/11398/1/Ling_Siong_Shi.jpg; http://studentsrepo.um.edu.my/11398/8/siong_shi.pdf; http://studentsrepo.um.edu.my/11398/

Video images have been widely used to extract relevant information for applications. One such application is the heart rate estimation using facial images from video sequences. A method using Canonical Component Analysis (CCA) is introduced to estimate the heart rate reading. In this study, two five-second video is used. The starting points of the two 5-second video frames are not the same. One of them starts one second later. It is assumed that these two 5-second video frames will contain the heart rate signals that are strongly correlated to each other and the random artifacts and noises appearing in both video frames are not correlated to each other. Therefore the desired heart rate will remain unchanged which have the ability to give maximum correlation values. The obtained eigenvalues and eigenvectors with maximum correlation values now can be used to find the desired original sources. By examining the most correlated signal within these two video frames using Canonical Component Analysis (CCA), the heart rate can be estimated. The identified heart signal from the result of CCA is then passed to a bandpass filter (0.8-4Hz) followed by Fast Fourier Transform to obtain the heart rate. Two experiments which involve increasing and decreasing dynamic heart rate were conducted. For both experiments, eight subjects took part in a cycling activity and their heart rates varied from 70 to 157 beats per minute (BPM). The results from both methods showed that acceptable average error rates of less than 3.70 BPM are observed between the actual and the estimated heart rates. In addition to that, a study was carried out to determine the minimum distance between the subject and video camera with fixed video duration. The results from this experiment showed acceptable error rates are observed when the distance between them is less than 130cm. In the last experiment, the effects of varying video duration but with fixed distance are carried out. Acceptable error rates are observed between the actual and estimated video duration greater than 6 seconds.