Closed chennuo0125-HIT closed 2 years ago
question1: the first pair of values. The second one is the RMS of the deviation I believe. question2: try to visualize it with open3d for an example. You can plot your cameras as frame with 0X, 0Y, 0Z axis. This will help you to estimate the result. But you need to transform T_2_1 and other T's accordingly.
so @ishipachev, 0.000017 +- 1.969469
is the reprojection error and 0.000003 +- 1.554876
is the RMS of the deviation?
@alwynmathew No. [0.000017, 0.000003] -- is the mean reprojection error (not sure if I'm correct) and [+- 1.969469, +- 1.554876] are RMS
@ishipachev why do we have two values as reprojection error [0.000017, 0.000003]
? How can the mean reprojection error in pixels be this low?
@goldbattle is this the correct way to interpret the calibration result in kalibr?
The error is being printed here if you look at the code: https://github.com/ethz-asl/kalibr/blob/438f7545a303a8c8bf0699623879f149a9fec71e/aslam_offline_calibration/kalibr/python/kalibr_camera_calibration/CameraCalibrator.py#L599-L606
Thus we have the mean and its standard deviation for the x and y pixel coordinates.
reprojection error: [-0.000008, -0.000002] +- [0.780705, 0.612191]
Here we can see that the system is zero mean (which is good) with a standard deviation of 0.7.
The example Kalibr dataset has a STD of 0.33 and is zero mean.
Its 3 sigma is 1.01 which captures the majority of reprojection errors if you inspect the PDF.
reprojection error: [-0.000002, -0.000000] +- [0.336932, 0.283749]
Normally, we want 0.1-0.3 reprojection error if you do a really good calibration. Thus you probably should recollect / focus your camera / try to improve your calibration if you are getting over 1 pixel sigma. Note that this also is dependent on the resolution of your camera (a 4K camera might be ok with 1 pixel if the FOV of the camera is the same as a 480p camera...).
Hope this helps.
i calibrated extrincs between multi cameras with kalibr, and get following results:
question1: Is the reprojection error [0.000017, 0.000003] or [0.000017, 0.000003] + [1.969469, 1.554876]? question2: the result is ok ?