Calibration values from camera_info tool

Calibration values from camera_info tool

I would like to know what the meaning calibration numbers displayed by the camera_info tool, and the corresponding calibration model used. Is this documented somewhere? I cannot find it.

First, there is a set of 12 numbers near the top of the file just associated with "Camera":

Camera: 1 (0x0001)
    Serial: ...
    Model: VF0780
    TOF controller version: 9477 (0x2505)
    USB backend version: 201918467 (0xc090803)
    Calibration
        0.999974 0.001318 -0.007029 0.001203 -0.999865 -0.016404 0.007050 -0.016395 0.999841 0.024492 -0.000508 -0.000863
    Capacities: Color | Depth | Audio

What do these 12 numbers mean? Also, is there any information about the relative pose difference between the color and depth cameras (translation and rotation)?

Next, there is a set of 11 numbers associated with the "Color Node":

   Color Node
        Revision: 2028 (0x07ec)
        PID: 16534 (0x4096)
        VID: 1054 (0x041e)
        Calibration
            640 480 583.078979 596.203003 320.000000 240.000000 0.022575 -0.162668 0.186138 0.000000 0.000000

I assume the first two numbers are just the width and height of the color images. Judging by my queries of PROPERTY_COLOR_FOCAL_LENGTH and PROPERTY_COLOR_PRINCIPAL_POINT, the third and fourth numbers of the x and y focal lengths and the fifth and sixth numbers are the (x,y) coordinates of the principal point. Is the principal point really at exactly half of the image dimensions? That seems suspicious. What are the rest of the numbers? Lens distortion model? If so, please provide a description of the model used.

Next, and probably of most interest to me, is the "Depth Node":

    Depth Node
        Revision: 2028 (0x07ec)
        PID: 16534 (0x4096)
        VID: 1054 (0x041e)
        Calibration
            320 240 224.501999 230.494003 160.000000 120.000000 -0.170103 0.144064 -0.047699 0.000000 0.000000

Judging by my queries of PROPERTY_DEPTH_FOCAL_LENGTH and PROPERTY_DEPTH_PRINCIPAL_POINT, the third and fourth numbers of the x and y focal lengths and the fifth and sixth numbers are the (x,y) coordinates of the principal point. Again, is the principal point really at exactly half of the image dimensions? And what do the remaining numbers mean? It would be great if someone can provide the models used.

Thanks.

-Scott

10 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

I will provide the details to you soon. Thanks!

The first calibration set is the transformation from depth 3D coords to color 3D coords. The 2 other sets are the intrinsic parameters of each cameras (image size, focal, principal point, distorsion). These parameters are used, for example, to create the UV map.

Cedric

Is there a way to get this calibration data at runtime ? I really need this data for my project and for the moment I must hard-code it (which make my application non redistribuable...)

Thanks
Cedric 

Except for the rotation data, the other data are identical from a camera to the other, which surely means that it is approximative calibration data.

Cedric

Calibration of the color node and depth node

// image width
// image height
// The focal length along the x axis expressed in pixel units
// The focal length along the y axis expressed in pixel units
// The central point along the x axis expressed in pixel units
// The central point along the y axis expressed in pixel units
// The first radial distortion coefficient
// The second radial distortion coefficient
// The third radial distortion coefficient
// The first tangential distortion coefficient
// The second tangential distortion coefficient

Thank you Cedric and David for the information.

David, I have the same need as Cedric. I would like to get this calibration data at runtime. Is there a way using the SDK to do that? Or are there plans to add that capability in the near future?

I also have a few other questions about interpreting the calibration numbers.

Are the radial and tangential distortion coefficients the same coefficients used in the model formulation in the Caltech calibration toolkit?

http://www.vision.caltech.edu/bouguetj/calib_doc/htmls/parameters.html

That is, are the radial distortion coefficients exactly the kc(1), kc(2), and kc(5) coefficients and the tangential coefficients kc(3) and kc(4) in the model described on the webpage above. There are slightly different calibration formulations, so I want to double check that I understand the meaning of the numbers.

For the first set of 12 numbers giving the transformation for depth 3D coordinates to color 3D coordinates, it seems clear that the first 9 numbers give the rotation matrix R and the last 3 numbers give the translation vector t=(tX,tY,tZ). Is R given row-wise? That is, are the first 3 numbers the first row of 3, followed by the second and third rows of R? Whether row or column-wise, the given R matrix has determinant equal to -1. So this transformation is actually a reflection. Is there a mistake here? I would expect a rotation matrix, not a reflection.

Thanks.

-Scott

Hi!

I'm currently dealing with the same problem than Scott.

Could it be possible to have some answers to their last questions?

Thank you.

Currently we do not have any plan todisclose the calibration parameter details. I willencourage your guys to use the Projection functions.

Hi, Scott,

Have you figure out the meaning of the calibration data, e.g. "Depth Node  320 240 224.501999 230.494003 160.000000 120.000000 -0.170103 0.144064 -0.047699 0.000000 0.000000"? They would be crucial if  they are the paremeters of focal length etc., since the depth data would be biased a lot if the distortion is not corrected.

As for Kinect, the coefficients are calibrated and given out in SDK, also there are quite a lot reseachers that found out ways to get more accurate depth results, e.g.  the TPAMI paper "Joint Depth and Color Camara Calibration with Distortion Correction" 2012.

If you are still interested in this issue, please contact me with wuzheming@gmail.com.

Best,

Zheming

Leave a Comment

Please sign in to add a comment. Not a member? Join today