Correct images with Camera corrector

Hi, I'm using the camera corrector application to correct a series of images that have barrel distortion using chessboards.

Following the steps below:

1.I attach the series of photos to be corrected with the different positions of the chessboard, without modifying the position of the camera.

2. I introduce the size of the different squares that make up the board. And frame that has a high distortion the image to process

3. I perform the processing of the different photos and press the calibration button

4. Detect the different points of the chessboard

5. And I hit the "show undistorted" button. To see the result of the correction.

Apparently the image is corrected. But using a photo correction program, I observed that this is not the case, that there are errors in its correction.

it is observed that the upper right corner is not fully corrected. I have carried out the same procedure with a smaller number of photos, with a different location of the chessboard, etc. I am somewhat confused. What could I do to completely correct the image?

Answer 1

[Image Analyst]

Looks pretty good to me. What sort of problems is it creating for you? Why do you need more accuracy? The distortion is probably computed as a regression through the detected corner points, and since it's a regression may not have subpixel accuracy for all points in the image. How much off is it and what problem does that cause in your analysis of the scene?

Your camera is also not pointed at the center of the scene and probably does not have the optic axis normal to the plane of the gray frame, thus causing skew/shear in the corrected gray frame. Is it super important that the gray frame be perfectly rectangular? If so, why? Try adjusting your camera so that the camera is at the center and perpendicular to the plane of the scene.

It might also improve if you made your checkerboard larger and put it near the center of the scene with its center on the optic axis.

It looks like you're using a very short focal length lens and are close to the scene, thus causing the fish eye (barrel) distortion. You could reduce the distortion in the first place by using a longer focal length lens, or course this would mean moving the camera farther away to get the same field of view, but maybe that's an option.

[Pedrp Perez Serrano]

I must have the image fully corrected inside the rectangle that makes up the test channel. Since later inside it circulates a flow of water that is digitized, if it is not totally corrected, the originated coordinates are not totally the real ones. Therefore the gray frame must be fully corrected.

Regarding the creation of a larger chessboard, it is something that I have considered. But I wanted to have an opinion about the size of this one. And regarding its placement in the center of the image, in the set of images taken with different positions, several of them have the position of the center of the image. So I don't think that was a problem.

In summary, I will create a larger chessboard, put the camera in a position further away and centered on the correction zone, and see if the results of the corrections change. What if I have discovered that the correction error always falls on the same corner. Thank you very much for your time and your ideas.

[Image Analyst]

Will the software work if there are multiple checkerboards, like one on the left, right, and top middle?

[Pedrp Perez Serrano]

Yes, each panel in a corner and in the center. In all possible positions, I attach all the images used for the "image correction" :

1)

2)

3)

4)

5)

6)

7)

8)

9)

10)

11)

12)

If you want, I can send you all these photos to an email with the parameters that I use. And compare it with yours processed. Thank you!!

[Image Analyst]

Your answer said, in essence, "I need it more accurate because I need it more accurate." You are never going to have it be 100% accurate, ever. So what you have to decide is if it's accurate enough for what you want to do. For example, let's say that the coordinates were biased 1 mm higher. But you want to find the upper level of 10 different fluids flowing across to see how that varies with fluid viscosity (or oscillation frequency or whatever you're measuring). Can you use those coordinates to compare the relative heights of the fluids even though they all read 1 mm higher that the "true" level? Sure you can.

I meant more than one checkerboard in a single photo. And I don't have any code, nor am I going to have time to develop it for you, since it seems like more than a 5 minute job.

And it looks like you haven't had time yet to back up the camera and use a longer focal length lens. I'm not sure that you're measuring in that photo (maybe the white line above the hump) but I'm pretty sure you can vastly improve the photo with better background (perhaps black velvet?) and better lighting (proper size and coming from an optimal angle).

[Pedrp Perez Serrano]

Thank you very much for your reply. In these photos I am not measuring anything yet, first install the panels to try to correct the image so that later I can measure the height of the fluid. Since if I install the panels at the same time that the fluid passes, it does not let me process it correctly.

I have decided to install larger panels to cover more surface of the image and also try to focus the camera on the center of it.