Estimating User Heading for phone in Pocket

Hello All,

It was painful to realise that magnetometer is NEVER reliable when
estimating user heading.

So, I ma using iPhone 6 for my testing where, I can access: TrueHeading,
MagneticHeading and Quaternions.

TrueHeading and MagneticHeading have huge variations for the case when the
user is walking with phone in hand. However, when I convert the
Quaternions to Euler and use the Yaw as Heading, it provides a much better
heading and user Trajectory.

So I thought to stick to this method where I convert the Quaternions to
Euler and use the Yaw (rotation around z-Axis, which is coming out of the
screen).

The problem I have is, when the user puts the phone in pocket (the screen
of the phone might face either inside or outside) and then the value of
heading is not as expected. The trajectory however, is "more or less
accurate". 

Any idea how I could solve this problem of estimating decent heading when
the phone is in pocket?

Best Regards

sia
---------------------------------------
Posted through http://www.DSPRelated.com