Failed to obtain orientation prior about kalibr HOT 5 CLOSED

Hi,

the orientation prior is computed by fitting an orientation curve to measurements provided by the camera and the angular rates measured by the gyroscopes. If the intrinsic calibration of the camera is reasonably accurate and there are no major scaling errors or large temporal offsets in the gyroscope measurements, this works reliably.
So these are all points to starts from: Please provide your report for the camera intrinsic calibration and please ensure that you do not have a scaling error on the rate measurements and that the timestamps are not vastly off. Of these issues, problems with timing are a good candidate to start with.

Cheers, Joern

from kalibr.

Any luck solving the problem? Can this issue be closed?

from kalibr.

Haven't solved it yet. I'm suspicious about a few things.

1. Somewhat asymmetrical camera calibration result, with straight lines on the right side of rectified image not being straight.
   

Two different camera calibration results, using ueye and ueye_cam packages.

height: 480
width: 752
distortion_model: plumb_bob
D: [-0.2935960992080, 0.08411669991252001, 0.00039760017211, -0.000271737787582, 0.0]
K: [420.5753013752, 0.0, 287.2677867092, 0.0, 420.0058738995, 239.086996746, 0.0, 0.0, 1.0]
R: [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0]
P: [336.567108154, 0.0, 272.100983125, 0.0, 0.0, 373.8412780, 238.612772, 0.0, 0.0, 0.0, 1.0, 0.0]

height: 480
width: 752
distortion_model: plumb_bob
D: [-0.27049, 0.058260000000000006, -2e-05, 8e-05, 0.0]
K: [418.78201, 0.0, 346.79708, 0.0, 418.75386, 241.70869, 0.0, 0.0, 1.0]
R: [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0]
P: [303.31152, 0.0, 329.04409, 0.0, 0.0, 375.34326, 241.62175, 0.0, 0.0, 0.0, 1.0, 0.0]

2)A very large ros system timestamp. When running mavros for example, this gives the TF_OLD_DATA warning. No idea how to fix it yet.

from kalibr.

Ok. I presume that these are two results for the same camera, in which case the difference in the estimation of the principle point is significantly too large. It further looks like that you obtained this calibration with the ros utility rather than with kalibr.
Possible problems are that you did not sufficiently sample the "checkerboard pose space" while calibrating. Personally, I collect multiple minutes of data while presenting the calibration target in various orientations and at different distances. A more principled approach can be found in "AprilCal: Assisted and repeatable camera calibration" by Richardson et al. . While they guide the user, kalibr takes a different approach by selecting views from the pool of data that contribute most to observing the calibration parameters. I think that the equidistant distortion is a better model of the effects present in most optics, a claim that seems to also be supported by the aforementioned publication.
Please record a sufficient amount of data and try the kalibr calibration function using the equidistant model to see whether it fixes your issues. I noticed that for this model, the distortion parameters vary widely for similar lenses while giving similar, low reprojection errors. Accordingly, please assess the quality of your calibration by the reprojection error on a test ste and the reproducibility of the focal length and principle point rather than the numerical values of the distortion parameters.

Regarding 2), this could quite possibly be the source of your problems, but it is not related to this software package. Please reconsider whether the problem that you are observing are rooted in our calibration toolbox or in your current system architecture.

Cheers!

from kalibr.

Hi, camera calibration result seems ok now, after running it with the kalibr tool and checking it with calibration validator. The cam-imu calibration also worked with a larger marker board. Thank you for support!

from kalibr.