The Effect of Noise on Contactless Heart Rate Measurement using Video Magnification

Abstract

Detecting heart rate via the non-contact method of Video Magnification (VM) is necessary when contact-based methods are overly cumbersome or not feasible, such as during a remote virtual consultation with a physician. The heart rate (HR) signal in video is best correlated with the miniscule change in skin colour associated with the presence and absence of capillary blood during the heart compression/recovery cycle. This change primarily affects the green colour channel in video at levels that are imperceptible to the human eye but can be detected by VM. However, there are many sources of noise that influence the performance of the VM algorithm, starting with the video capture process and ending with the algorithm parameters. In this paper, the VM algorithm performance in the absence and presence of noise was investigated under controlled conditions through the creation of artificial videos with set parameters to better assess algorithm performance. It was found that in the absence of noise, the VM algorithm can accurately detect the simulated HR frequency with a signal amplitude as small as ±1 quantization level. Moreover, it was also found that the algorithm can detect the simulated HR frequency with a signal amplitude as small as ±0.5 quantization level in the presence of a small level of noise. Lastly, it was found that although algorithm performance degraded with increased noise, the simulated HR results could be found in signals as low as ±1 quantization level even with noise at a power equivalent to ±32 quantization levels.