Determining a suitable, adaptive region of interest (ROI) automatically for extracting information related to cardiac activity (signal-ROI/S-ROI), and another containing information on ambient light-fluctuation (background-ROI/B-ROI, as close as possible to… Click to show full abstract
Determining a suitable, adaptive region of interest (ROI) automatically for extracting information related to cardiac activity (signal-ROI/S-ROI), and another containing information on ambient light-fluctuation (background-ROI/B-ROI, as close as possible to the signal-ROI), and robust signal processing are important in webcam based heart-rate (HR) estimation – in real life situations. We describe a novel method of automatically determining both the ROIs. The forehead is the candidate for the S-ROI, due to its uniformity and minimum vulnerability for deformation. We first identify the skin-pixels within the face-region detected by the Viola-Jones (VJ) algorithm. The forehead-region, and a uniform sub-rectangle within it not containing hair – determined by using variance as a measure – yields the S-ROI. The B-ROI – consisting of 3 rectangles – each of the same size as that of S-ROI, at the two sides and the top of the VJ-rectangle – is used to generate a reference signal for an adaptive noise-cancellation scheme. The situation arising from (possibly simultaneous) facial expressions deforming the S-ROI, is addressed – by extracting the phase sequence associated with the analytic representation of the signal. Experiments conducted with 21 healthy subjects, using this novel array of techniques, have produced good correspondence with the ground truth obtained from a standard finger-pulse transducer – as reflected by the Bland-Altman plot.
               
Click one of the above tabs to view related content.