Heart Rate Detection using Video Magnification: Impact of Algorithmic Parameters and Noise

Screening people for illnesses in public spaces using contactless techniques such as Video Magnification (VM) can be beneficial for both the crowd and the operators. One important vital sign to screen for is heart rate because it is indicative of the overall health of humans. In this thesis, first the impact of varying two input parameters (region of interest (ROI) size and window segment length) on VM is explored. Second, the impact of noise from three different sources (quantization, modern camera systems coupled with software enhancements, and light illumination) is explored. Results show that heart rate can be detected from very small ROIs, but that larger ROIs have higher accuracy. Larger window segment lengths are useful in the presence of motion. VM detects signals with amplitudes equal to one quantization level in 8-bit videos. VM is affected by modern camera systems, and it is more accurate with higher light illuminations.