I have had some entertainment downloading an app that just feeds the accelerometer data to the screen. I did it first as a diagnostic to see if the accelerometer was broken (it was on this machine), but then I have some fun just watching the data on the screen and trying to imagine how on earth you're supposed to extract whether the user is standing or not, or going up stairs, etc.
And I say that from a position of being educated in signal processing, not just staring slackjawed at all the numbers. It's a very noisy signal, even after processing, and while it's correlated with the values you're trying to get out of the data it's not necessarily a very strong correlation. There's a lot of headwinds against getting this data out of the accelerometer. It's a minor miracle that they are as accurate as they are.
I would expect it to be accurate enough to detect the changes: if it is accurate enough to detect stand up and move, it also should pick stood up, moved a bit, didn't stand down. I guess the profile of accelerometer/altimeter data of standing up and moving from sitting is different from stand up and moving without having been seated before.
And I say that from a position of being educated in signal processing, not just staring slackjawed at all the numbers. It's a very noisy signal, even after processing, and while it's correlated with the values you're trying to get out of the data it's not necessarily a very strong correlation. There's a lot of headwinds against getting this data out of the accelerometer. It's a minor miracle that they are as accurate as they are.