I'm wondering how you can get a non-drifting signal like here:

 

In the comments, the OP claims that only the quaternions are needed. They are indeed quite stable on my display, while the linear acceleration is very noise (in the magnitude of 0.1 m/s^2 error).

 

How would I do the integration based on quaternions? It sounds odd to me, as they are only giving orientation, right?

 

Any ideas?

 

best, ..h.h..

Ok, so the trick in that video is that it uses three IMUs, and assumes the first (top-most) is not moving but only rotating; so it has precise rotation for that, then extends a vector of known length from there, adds a second rotation, extends another known vector, adds the third rotation; thus only rotations are used, and this doesn't solve any real motion tracking problems.

 

Has anyone found good filtering techniques to minimise the drift problem of the accelerometer?