Heart Rate Estimation from Facial Video Sequences using Fast Independent Component Analysis

In this paper, a non-contact heart rate measurement method is proposed, which gives a comfortable physiological examination of cardiac pulse without the use of electrodes. This method is based on automated face tracking and blind source separation of the colour channels into separate components and used on colour video recordings of the human face. Here by optimizing non-Gaussianity and negentropy for the recovered signals, an a FastICA (Fast Independent component Analysis) algorithm is employed to extract independent components. For experimentation, COHFACE dataset is used consisting of 160 video's of 40 different people (28 males and 12 females). Heart rate estimated with FastICA is compared with heart rate measured using Finger blood volume pulse (BVP) sensor. This comparison is performed using Bland-Altman and correlation analysis. With proposed method low error rate is observed when compared with Independent Component Analysis (ICA) and other methods with same database.