Reducing Motion Impact on Video Magnification Using Wavelet Transform and Principal Component Analysis for Heart Rate Estimation

Abstract

In this paper, we developed a contactless method to estimate human subjects' heart rates using a magnification of video sequences. Existing methods based on the Eulerian video magnification (EVM) technique are severely affected by motion (i.e., object movements). The proposed method can consider both scenarios of still and moving subjects to magnify the skin color changes over time. In the proposed method, we employed a wavelet decomposition process along with a denoising layer based on principal component analysis (PCA). PCA smoothed the frames and reduced the noise due to high-frequency fluctuations caused by motion (i.e., unwanted source of color changes in the skin). Finally, the video data was magnified to extract the heart rate information. As a result, subtle changes caused by blood flow were made clearly visible to the naked eye. We compared the heart rate estimation results obtained through our proposed method with the results produced by linear EVM. The experimental results indicate an improvement in the heart rate estimation accuracy in the presence of motion using the proposed method.