This is a bit niche, but quite interesting to me.
I've picked up one of these new-fangled activity monitors, specifically a Garmin Vivoactive HR+. I already track all my "proper" exercise, but I was curious about what my daily casual activity was, and also my sleep quality. This particular model had a few features I knew I'd use (GPS for casual cycling, 24/7 HR monitoring for sleep/resting HR tracking, and I already use Garmin's backend for a lot of my data-gathering).
One thing I wanted to check on was the accuracy of the optical Heart Rate monitor. I'm used to using a chest strap, that picks up the bio-electrical pulses from the heart. Optical HR monitors use an LED to shine through the skin, and measure blood flow based on the scattering of light (there is a load of info here if you want to read more). Most reviews I've seen of this method have indicated a lack of accuracy, especially at the beginning of exercise (when blood may be away from the skin), at peak heart rate levels (when the blood is simply moving too fast to track), and during interval training (when the rate of flow is changing quickly). Add in the challenges of an imperfect light seal (ironically, you can't have the wrist strap on too tight, or it causes constrictions, and also starts to get a bit skanky...these things are supposed to be worn 24/7, unlike a heart rate strap), and movement of the sensor, and you have a tough environment.
I decided to run the optical sensor alongside a traditional chest strap on a simple indoor training session. The below graph shows the results. The session was;
7 minutes - gradual increase in effort as a warm up
3 minutes - recovery
40 minutes - steady "sweet spot" effort (sustainable effort, but hard enough that you can't talk)
10 minutes - gradual reduction in effort
The heart rate strap is in red
The optical heart rate sensor is in black
The horizontal axis is the pedal revolution count (most cycle-based monitoring systems record data every pedal stroke...I was turning a cadence of 90-95 for the entire session). The vertical axis is heart rate.
You can see that the first 10 minutes is all over the shop for the optical HR sensor. I had to extrapolate a lot of the data points, as it simply didn't take that many readings either (even though I had it in "Activity Mode"). Once I was into the "steady state" part of the session, it became a lot more accurate, though again you can see the lack of data sampling meant that it's reactions to any change are quite slow.
I might repeat this test when I do a "spikier" interval-based session...this is probably a best-case scenario for the optical HR sensor, with a long period of reasonably consistent heart the readings. The first 10 minutes, however, are atrocious, and I'm really surprised that it struggled that much to get a reading. It's clearly trying, as there is a high density of data points in this time period, but it repeatedly fails to accurately determine the heart rate.