Contactless vision-based pulse rate detection of Infants Under Neurological Examinations

In this paper, we propose a method for detecting variations in the Pulse Rate (PR) of infants undergoing the Hammersmith Infant Neurological Examinations (HINE) using video data. As in every other medical examination the measurement of the PR is critical to underpin the physiological state of living beings. During HINE, measuring the infant's PR is important as its variations against physical conditions, age and other factors must be studied and correlated against developmental scores. However, this becomes highly complicated with active infants where their movements often lead to inconsistent PR estimation. We propose the use of a non-linear dimensionality reduction technique, called Laplacian Eigenmap (LE), to uncover the pulse information encapsulated within the high dimensional visual manifold characterized by normalized RGB feature vectors. Furthermore, low-level image filtering is applied to accurately detect PR within a chosen region-of-interest (ROI) from different parts of the infant's body. For validation and analysis, a set of 14 video sequences of infants undergoing five important tests of HINE have been chosen. Experimental results suggest that a bi-parametrized combination of color features from the RG and GB channels provide more valuable information in comparison to the RB and RGB channels. Results have demonstrated that this contactless method of PR detection has promising prospects for its future use in other clinical examinations of infants.