puzzlepaint
commented
3 years ago In general, computing the relative pose between two cameras is not something that is related to the camera calibration project in this repository here, so I would consider it off-topic. As mentioned in #52, I cannot provide general Computer Vision support here, I am sorry.
Regarding the difficult convergence of the calibration with multiple cameras (but success with individual cameras), some ideas for possible reasons for that might be:
- Low overlap in the field-of-view of the cameras might perhaps cause issues. If possible, perhaps adding a third camera temporarily (only during calibration) to get higher overlap might help.
- The camera model might not be ideal for the camera. I believe that this becomes better visible when calibrating multiple cameras together, as I think that with only a single camera (and the calibration pattern covering a limited part of the camera image only), the optimization process of the calibration may have some room to compensate shortcomings of the camera model by (incorrect) adjustments to the camera pose. When multiple cameras are calibrated that have a fixed relative pose, then there is less room for the optimization process to do such compensations. So, in case you did not use the (most generic) non-central generic model, it might be worth trying whether this gives a better result.
- The cameras might not be in a 100% fixed relative pose to each other, violating the assumption of the fixed relative pose.
Dear @puzzlepaint , I want to calibrate a stereo camera. I tried to calibrate them together, but the program was difficult to converge. But when I calibrate them separately, they converge well. The most important thing is that I want to know the RT between two cameras. I can calculate R through the calibration results, but I don't know how to calculate T. Can you give me some ideas to solve the problem?Thanks!