Tips for performance?

[y22ma]

It would be much appreciated if a section on how to get the best performance from this package can be added.

Currently, both stereo_vo and mono_vo either crashes because there are not sufficient correspondences, or its pose estimates are diverging wildly if I lower the rejection thresholds.

[hengli]

The rejection thresholds are already low enough. It is important to have a accurate calibration for the monocular/stereo camera in use. Perhaps you can describe your situation more specifically so that I can have a better idea of why too few correspondences are being detected.

[y22ma]

After calibration using the stereo_calibration_node, I got stereo_vo working fairly consistently. The reprojection error is around 0.45 for both cameras.

With this calibration, however, mono_vo is not really capturing the camera translation. When I move the camera in pure sideway translation, the pose estimates output indicates a rotation without much translation at all. Just wondering what would likely cause that?