In the past, we have covered camera calibration and the effects of autofocus. Now that calibration is easier, it looks like there may be some issues with the positional repeatability of the auto focus lens mechanism. Initial results seem to indicate that every time the focus of the camera changes not only does the focal length change but it looks like the orientation of the primary axis and position of the center of the lens relative to the principal point also change.
A little more testing is need but the long and the short of it is that the Logitech Quickcam 9000 may not be useful for many computer vision applications since you may not be able to rectify the image consistently unless you recalibrate the camera every time the focus changes. A lookup table won't work since you have no way of determining how the orientation of the lens has changed.
Oh and that auto focus mechanism, it's an electromagnet. As we have previously noted, magnetic fields have some implications for inertial navigation. One thought I had today was that often the intensity of the magnetic field is ignored because generally the strength of the Earth's magnetic field is constant and you are only looking to determine the orientation relative to magnetic north.
Perhaps the covariance for the Kalman filter could be weighted based on the magnetic field strength relative to the nominal field strength of the Earth. That way, if there is magnetic interference, the sensor readings from the magnetometer will have less influence on the output. Maybe someone else has researched this problem.
Key features of Droiko
17 hours ago