[colanderman]

Except that integrating any real world signal tends to introduce drift due to noise in the input signal. If you have a smartphone, try writing an app that does exactly what you propose; you'll soon find that it's impossible to get the "speed" to zero out. (Either that or you have to filter the data, which will cause other fun errors like your speed drifting toward zero.)

[Bud]

I thought about this as a complication, but my assumption was that this would be a solvable technical problem. Is it really not feasible to add an accelerometer to the GPS data (which does include altitude), and end up with at least somewhat more precise data?

[zb]

You can feed the accelerometer data into the Kalman filter as well. However, since it's much more noisy than the GPS data, it won't give you any real benefit.

Where people would like this is to do dead reckoning when a GPS position isn't available (e.g. in tunnels). However, it turns out that it is completely useless for that, for the exact reasons the GP describes.