EKF2: vision attitude error filter

[bresch]

Solved Problem

When the vision frame is not NED, we need to rotate the EV data by the attitude difference between the two frames. The issue is that any noise in that difference will be transferred on the EV position and velocity measurements.
This can in some extreme case cause a constant reset of the estimator as the data can often fail the innovation check.

Solution

Since the attitude error is almost constant (really slow drift), it can be heavily low-passed to remove most of the noise.

Changelog Entry

For release notes:

Add vision attitude error filter
New parameter: -
Documentation: -

Alternatives

estimate the attitude error

Test coverage

unit and sitl tests