User's Manual
Document MV0319P.N
© Xsens Technologies B.V.
MVN User Manual
122
19.9 XKF-HM
Xsens Motion tracking works on the basis of Kalman Filtering. In its simplest form, it combines the
different information sources to calculate the desired parameters. This has an impact on how the
systems should be used. The more historically correct data has been measured (in good magnetic
environments for example and with enough physical movement of the motion trackers) the more robust
the filter can be. For this reason we advise to calibrate in a magnetically sound environment (See Section
3.4.1) and to “warm-up” the filters by moving casually for a short time, to build up this history (See below
for settling time details).
With MVN Studio 4.0 and onwards, Xsens has created a special filter dedicated to accurately measuring
human motion. This filter is called XKF-HM (Xsens Kalman Filter for Human Movement) and is based
on a number of assumptions that humans will move with varying degree of speed, high and low
accelerations, while assuming that the duration of accelerations will not be longer than is humanly
possible.
The XKF-HM filter works well in a variety of environments and dynamics. This means that measurements
can take place of movement that is semi-static up to high dynamics as well as being possible to stay in
magnetically challenging environments for longer. This has been made possible due to the improved
signal pipeline.
An important aspect of using any measurement apparatus is the scale of the sensor. Often, for example
with a temperature sensor (thermometer!) when the range is exceeded, the sensor either breaks or
simply can no longer measure outside of the defined range. With Xsens sensors, when the data exceeds
the limits there is a potential of this influencing the calculated orientation and could last more than a few
sample frames due to the history build up. However with the new algorithms this is completely prevented.
Clipping is of course possible, when the range of the component is exceeded. However, the influence
of this clipping is negligible. For the component itself, when the movement is back in range the
measurement continues as normal and the calculated orientation returns to an accurate state
immediately. It is also possible to observe when clipping occurred during recording, so that if this data
is required it can be ignored. Furthermore in future releases this will be possible to reprocess and will
have no effect (since both historical and future data are known in a recorded file).
Other remarks: Some amount of correction for the changing magnetic field is present in the new XKF,
this is due to three things: the improved signal pipeline mentioned above, a new model that is used for
modelling magnetic field behavior and compensation and that there is also some recalibration of the
magnetic field taking place in this algorithm. This means that when a new magnetic environment is
present, the filter understands that this is new and temporary and can compensate for this for a longer
time.
19.9.1 Height tracking
Due to the many improvements, from signal processing to how data is handled in MVN Studio including
foot contact detection, it is now possible to track height changes of a subject e.g. walking upstairs or up
a steep slope while wearing the system. Please note that height changes due to e.g. changing floor level
in an elevator will not be detected, since there is no relative change in motion of the body segments, in
this case, it will appear that the subject stands still and continues on a level plane.
19.9.2 XKF-HM Settling time
Settling time is the time needed for the filters to “warm up”. When they are warmed up, information like
gyroscope and magnetometer biases can be estimated and uncertainties minimized.
Tests have shown that the best practice for achieving a good settling time is to remain still for 30 seconds
after the initial calibration (N or T pose).