If the velocity is slower than FAST then we look back up to 30 seconds and
calculate the velocity for the past 30 seconds instead.
For the first version I'm not doing the average of the changes but simply
the change from beginning to end.
The alternative would be to do another triangle smoothing or something
like that - but as we don't know how many samples we have in the 30 second
window, it's a little harder here.