In the microsoft method, it would be reasonably easy to approximate their SFM step with info from the gyro sensor. From their paper, SFM seems to be one of the most expensive steps.