The camera uses two cameras and reflective infrared markers to compute 3D positions. For recognized marker configurations, the camera computes a translation and an orientation (quaternion). For stray markers, only the position is returned. The camera sends updates at 20-60 Hz.
To compute the depth (z-value) of a marker, the camera uses two noisy images. That means the noise in the z-dimension is different (and larger) than the noise in x and y. The noise increases with distance. For recognized configurations (translation+orientation), the noise distribution is pretty unclear.
How would you go about filtering this kind of data? We're mainly interested in getting rid of the jitter which is most noticeable when the trackers are just lying around or moving very slowly.