questions about feature tracking about open_vins

[goldbattle]

(1) So Q will have the orthogonal directions to the pFinA, thus this is basically doing the dot product for each two orthogonal unit vectors with p_CiinA. More simply, this is projecting p_CiinA onto the plane whose normal is defined by pFinA. This distance norm is the "parallax" in this created plane. This is different from just taking the norm distance difference ||pCiinA|| as we are only interested in the parallax between frames, since moving towards a feature does not add any information.

(2) So if you don't enforce the stereo tracks you get the "duo" camera tracking setup. I recommend you checkout this paper by Mrinal Paul, which talks about what you mention. Basically, you don't gain any accuracy in your triangulation step and only gain in getting more measurements as compared to stereo. We also explored this in our multi-camera paper here.

[pzhu008]

Hi Patrick,
It helps a lot. Thanks.