Weird Result of Camera Intrinsic and Extrinsic Parameter

[Xiang Li]

Hi,

I don't know if this group is still active, but it is worth a try. I downloaded version one and it is running good. I got projection matrices from three camera. But the results were weird to me:

Here is one example:

K11 = 17446.294561         R11 = -0.665322           t11 = -0.127741 

K12 = -1.237375                R12 = 0.662724           t21 = -0.123786 

K13 = -4415.947840          R13 = -0.343720          t31 = -6.745850 

K21 = 0.000000                 R21 = 0.655082 

K22 = -17444.410370        R22 = 0.739070 

K23 = -2650.773560          R23 = 0.156984 

K31 = 0.000000                 R31 = 0.358070 

K32 = -0.000000                R32 = -0.120720 

K33 = -1.000000                R33 = -0.925858 

1. I am wondering why the intrinsic parameters are all negative? So the coordinate system it use left hand system? If so how can I converted to right hand system.

2. I have a known scales number for the points on images. How can I pass the scales to the calibration system? Which file I should use?

3. If everything is good, how can I verify my result?

And finally, think you for your excellent system. It is really help!

Sincerely,

Xiang Li

[Andrew Straw]

Dear Xiang,

Answers inline below.

> On 8 Nov 2019, at 4:04 AM, Xiang Li <lixia...@gmail.com> wrote:
>
> Hi,
>
> I don't know if this group is still active, but it is worth a try. I downloaded version one and it is running good. I got projection matrices from three camera. But the results were weird to me:
>
> Here is one example:
> K11 = 17446.294561 R11 = -0.665322 t11 = -0.127741
> K12 = -1.237375 R12 = 0.662724 t21 = -0.123786
> K13 = -4415.947840 R13 = -0.343720 t31 = -6.745850
> K21 = 0.000000 R21 = 0.655082
> K22 = -17444.410370 R22 = 0.739070
> K23 = -2650.773560 R23 = 0.156984
> K31 = 0.000000 R31 = 0.358070
> K32 = -0.000000 R32 = -0.120720
> K33 = -1.000000 R33 = -0.925858
>
> 1. I am wondering why the intrinsic parameters are all negative? So the coordinate system it use left hand system? If so how can I converted to right hand system.

I don’t know the fundamental reason. To deal with this, you can use the get_flipped_camera() method in pymvg. https://github.com/strawlab/pymvg/blob/6e48c3b3ccbc374dc928476891753d6ad4ff6521/pymvg/camera_model.py#L743

> 2. I have a known scales number for the points on images. How can I pass the scales to the calibration system? Which file I should use?

MultiCamSelfCal only returns calibrations up to a scale, rotation, and translation. You need to fit these parameters. Again in pymvg, there is an implementation: estsimt() https://github.com/strawlab/pymvg/blob/6e48c3b3ccbc374dc928476891753d6ad4ff6521/pymvg/align.py#L3

> 3. If everything is good, how can I verify my result?

You can always check that reprojection errors are low (e.g. sub-pixel). Additionally, if you have known coordinates for some of your points, verify that the reconstructed points are low error.

> And finally, think you for your excellent system. It is really help!

Yes, thanks to Svoboda and the original authors of MCSC, all the contributors in the meantime!

[Xiang Li]

Thank you Dr. Straw,

I have another question: Is the cameras in the same coordinate frame after running the gocal.m?

Best,

Xiang Li

[Andrew Straw]

Dear Xiang,

I am not sure what you mean. The script you mention, gocal.m, takes the 2d coordinates and produces camera calibrations. It is the main driver script for all of the MCSC.

Best,

Andrew

--
You received this message because you are subscribed to the Google Groups "MultiCamSelfCal" group.
To unsubscribe from this group and stop receiving emails from it, send an email to multicamselfc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/multicamselfcal/ec9f9539-6cbc-4513-ae82-2bd27fbd9e77%40googlegroups.com.

[Xiang Li]

Sorry for the confusion that I made Dr. Straw.

I meant are those calibrated camera in the same coordinate system? Like Are those both at world coordinates or they are both at the coordinates with reference to most left camera?

Thank you again for your reply. I am so appreciate your time for answering questions.

Best,

Xiang Li

To unsubscribe from this group and stop receiving emails from it, send an email to multica...@googlegroups.com.

[GAO ME]

您好， 

       我现在在尝试使用svoboda开发的相机自标定程序，我看到你在github上面进行了讨论，是中国名字，所以想向你请教一下。

       我现在使用im2poingts可以准确的识别到每张图片led灯的坐标，然后用gocal.m进行相机标定，得到的结果与我实际摆放的位置完全不一样，重投影误差只有0.26，现在我不知道该如何解决了，不知是否能得到你的帮助。

        谢谢。

Hello,

I am currently trying to use the camera self calibration program developed by SVOBODA. I saw that you discussed it on GitHub, which is a Chinese name, so I would like to ask for your advice.

I can now accurately recognize the coordinates of each LED light in the image using im2points, and then calibrate the camera using gocal. m. The result obtained is completely different from the actual position I placed it in, with a reprojection error of only 0.26. I don't know how to solve this now, and I wonder if I can get your help.

Thank you.

[Andrew Straw]

The results are only accurate up to a homography. Is this the issue? To get real coordinates, you will need another step such as aligning with some known reference coordinates.

[GAO ME]

Thank you for your reply

I probably didn't express myself clearly. I arranged the five cameras in a circular array, as shown in Figure 1, but the calculated relative positions of the cameras (as shown in Figure 2) were completely inconsistent with the actual situation.