MIT-SPARK/Kimera-VIO

Two stereo cameras

pooja-kabra opened this issue · 1 comments

Hi, if my setup has two stereo pairs, what changes should I expect to make in the system to be able to use the extra stereo?

We did some work to support multi-camera setups (see here), however this work will not be open-sourced. The changes are not trivial. The fastest way to do it would be to run two frontend modules in parallel, and pass all measurements to the backend. You will have to keep track of which feature tracks are from which frontend, though the final optimization will include all of the data.

I can say that the primary improvement gained from multi-camera setups was in the monocular case, as the scale estimation was improved. However, with stereo we were unable to get much better than with just one stereo camera, since extrinsic calibration was difficult and the error from poor calibration resulted in a lot of outliers in the final factor-graph optimization.