ensure very good even lighting on the board to help with corner detection and sub-pixel optimization.move the board closer/further from cameras to fill the volume occupied by the subjects.But do twist the board slightly (up to about 30 degrees) so that it is not always co-planar with the cameras’ sensors.portrait to landscape, or reverse, depending on how you define rows vs columns) more than about 30 degrees.
Make sure to not twist the board (i.e.make sure the chessboard is large enough to fill most of the view when it is held the same distance from the cameras as the subjects will be.However, since the lenses are relatively close to each other (depending on how far the object of interest is from the cameras), I would expect that the calibration would be very sensitive to how you are running the calibration. Deeplabcut 3d, or GitHub - gupsegunes/ZedCalibration: ZED stereo camera calibration with python) as if they were independent cameras. I’ve never used them, but ZED cameras seem to have factory calibration, or can be calibrated using standard methods (e.g. The suggestion for half the distance is just a general guideline to offer a starting point, but you can get good calibrations from many other configurations.
0 kommentar(er)
