# Quaternion covariance matrix for imu_filter.

Hi everyone,

I really hope that someone can help me as I'm quite stuck. I'm using imu_filter to apply a Kalman filter to my visual odometry / IMU measurements. My only problem is: imu_filter expects rotations as quaternions, and hence also wants the covariance matrix to be based on quaternions.

From my odometry, however, I only get Euler angles. I've been trying a lot to at least get an estimate for the quaternion variances, but so far I haven't been very successful. Googling, I found one very short (actually just 2 pages) technical report of which I think might cover this problem: Link

But I'm absolutely unable to understand it. Does anyone have an idea for a simple way to at least get a rough estimate of the variances? I don't want to use a static parameterization, as the variances vary quite a bit. I'd really appreciate any help.

Thanks, Konstantin