how would that work? I can understand how pulse is detected, as you are measuring the differences from second to another and finding a pattern. But how would you know what correlates to a particular level of oxygen in a particular person?
See http://en.wikipedia.org/wiki/Pulse_oximetry for more information, but basically you measure the difference in absorbance between two specific wavelengths of light (one is near-IR and might not be captured by a cellphone's camera) over time. The difference tells you how saturated your hemoglobin is (most of which has oxygen bound to it).
Pulse oximetry is one of the more reliable/least invasive/cheapest methods of determining pulse rate. Additionally, there's a massive market opportunity for extremely cheap PulseOx sensors that keep a data history: busy hospital emergency rooms would love to slap such a sensor onto the finger of every patient that comes through the door (and subsequently has to sit and wait). When a patient crashes while sitting in the waiting room the pulse ox sensor would be able to give the nurses and doctors a little bit of information about what happened that lead to the crash.
From what I've read it simply takes the ratio of absorbance from the systole and diastole and considers that a stable non-linear estimator of %SpO2. The device itself needs to be calibrated to get the non-linear fit in place, but each individual is assumed to be (in gross measure) similar enough to get meaning out of it.
Of course, there's going to be huge variation across finger types, blood volumes, lighting conditions, placement. I wouldn't consider them to be accurate much more than those impedance-based body fat percentage computers. In both cases, huge discrepancies from healthy values are still telling despite small-scale noise.
