calibration result far from "correct" about industrial_calibration

[xiqiaosang]

Dear @drchrislewis :
Thank you very much for your last reply. Now I followed your instructions, to put two cameras far away .etc. The result now is much better: Camera 1 can observe the target, and the calibration result for camera 1 seems to be accurate. However, camera 2 can not recognize the same circles-target, the target we used is modified_circle_5x5, and the following errors show up:

[ERROR] [1447663919.912224744]: intermediate frame from world_frame to world_frame
[ERROR] [1447663919.912477029]: roi = 0 640 0 480
[ERROR] [1447663919.912520319]: roi = 0 640 0 480
[ERROR] [1447663920.291806547]: cameras triggered
[ERROR] [1447663920.292018266]: waiting for camera observations to finish
[ERROR] [1447663920.292073112]: observations done
[ERROR] [1447663920.292141341]: extracting observations
[ERROR] [1447663920.564058955]: intermediate frame pose = 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
[ INFO] [1447663920.564388375]: Processing 25 Observations
[ERROR] [1447663920.564686868]: waiting for camera observations to finish
[ERROR] [1447663920.564804600]: observations done
[ERROR] [1447663920.564922810]: extracting observations
[ERROR] [1447663920.700662482]: couldn't find
[ INFO] [1447663920.701382036]: Processing 0 Observations

And I find out that the final result for circles-target and camera 2 in world frame remains the same as their initial values, no changes. I guess the reason behind this is due to camera 2's failure of recognizing the circles-target.
How can I solve this problem? Wish to get your opinion.

Best regards,
xiqiaosang

[drchrislewis]

There are several things one can do to get the opencv circle finder to
detect the grid of circles. The easiest is to move the grid so that it is
detected. If you've run ROS's calibration, you will note that when the grid
is too close, too far, or too oblique, or the lighting/exposure is not
good, that the target is not found. We use the exact same subroutine to
find the grid. The second thing you can do is add a region of interest
which narrows the search. Often this works, but I'm not sure why. Next, you
can increase the size of the grid, either more circles, or larger
diameter/spacing. Next there are lots of parameters for the grid finder. I
use the defaults, but if you modify the code to adjust these you might get
it to find the target. Oh, also when the target is on the verge of being
located, meaning it finds it sometimes, but not always, you can simply add
repeated observations. Have it try 10 times. There is some image noise, so
the images won't be exactly identical. This last idea rarely works.

On Mon, Nov 16, 2015 at 2:59 AM, xiqiaosang [email protected]
wrote:

Dear @drchrislewis https://github.com/drchrislewis :
Thank you very much for your last reply. Now I followed your instructions,
to put two cameras far away .etc. The result now is much better: Camera 1
can observe the target, and the calibration result for camera 1 seems to be
accurate. However, camera 2 can not recognize the same circles-target, the
target we used is modified_circle_5x5, and the following errors show up:

[ERROR] [1447663919.912224744]: intermediate frame from world_frame to
world_frame
[ERROR] [1447663919.912477029]: roi = 0 640 0 480
[ERROR] [1447663919.912520319]: roi = 0 640 0 480
[ERROR] [1447663920.291806547]: cameras triggered
[ERROR] [1447663920.292018266]: waiting for camera observations to finish
[ERROR] [1447663920.292073112]: observations done
[ERROR] [1447663920.292141341]: extracting observations
[ERROR] [1447663920.564058955]: intermediate frame pose = 0.000000
0.000000 0.000000 0.000000 0.000000 0.000000
[ INFO] [1447663920.564388375]: Processing 25 Observations
[ERROR] [1447663920.564686868]: waiting for camera observations to finish
[ERROR] [1447663920.564804600]: observations done
[ERROR] [1447663920.564922810]: extracting observations
[ERROR] [1447663920.700662482]: couldn't find
[ INFO] [1447663920.701382036]: Processing 0 Observations

And I find out that the final result for circles-target and camera 2 in
world frame remains the same as their initial values, no changes. I guess
the reason behind this is due to camera 2's failure of recognizing the
circles-target.
How can I solve this problem? Wish to get your opinion.

Best regards,
xiqiaosang

—
Reply to this email directly or view it on GitHub
#63 (comment)
.