It shows how to see heart beats in ordinary video footage
The Massachusetts Institute of Technology has struck again. Or rather, scientists from its Computer Science and Artificial Intelligence Lab (CSAIL) did. They have given normal video cameras the ability to check your heart.And we aren't being metaphorical at all. MIT scientists combined good old mathematical analysis with some filtering algorithms and averaging processes.
The result was a means to analyze skin color changes in order to figure out a person's pulse, the heart rate, and if their blood pumps are behaving in any way erratically.
The heart rate variability (HRV, the moment-to-moment deviations from constancy) can be used to diagnose potential heart issues.
The margin of error for their method of figuring out the HRV based on normal video is, or will be, of a few beats per minute.
Thus, smartphone camera, security cameras, demo camera setups and, of course, normal photo and video capture devices could eventually gain the ability to detect heart problems even in people who didn't even consider the possibility of having such health issues.
The CSAIL researchers have been having trouble with extracting interbeat variability though, because the noise introduced by those variable beats can confound the real HRV signal.
Motion and colorimetric imaging will have to be combined before the error margin drops below 2%. Anything higher and the technology can't be trusted.
And since signal extraction depends on the hardware (unlike the finesse that can be improved through software optimizations), we imagine that the accuracy increases the better the camera becomes.
Thus, since smartphone cameras have been deemed viable, the sensors of actual cameras, from the pocketable ones to large, professional equipment, will help even more.
What remains is to take other factors into consideration. Pulse, blood oxygenation, pupil dilation and skin resistance are a good start, but thermal analysis, spectrographic detection of components in sweat, and finer-grained spatial detection could help a lot.