How do I obtain camera Extrinsic parameters (matrices R and t) without using a checkerboard?

My ultimate goal is to produce a Projection Matrix file in order to be used for creating a 3D photo for a cylindrical object by combining several 2D digital images taken from different angles.

I used a camera calibration toolbox (available at: http://www.vision.caltech.edu/bouguetj/calib_doc/htmls/example.html) in order to obtain the camera Intrinsic parameters (matrix K) based on checkerboard images. Now, I'm stuck how to obtain the Extrinsic parameters (matrices R and t) for the images I took for my object?! I don't have a room to let the checkerboard appear next to my object in the images I want to take.

Please help! Thank you.

P.S. My background is civil engineering, I'm pretty new to computer vision area.

Answer 1

[Dima Lisin]

To compute the extrinsics, you need to somehow define a world coordinate system, because the R and t must be relative to something. Having a checkerboard in the scene is very convenient. If you do not have one, then you need to have some points or markers in the scene, which you can reliably detect in the image, and whose world coordinates you know. For example, you may have a picture on the wall which you can somehow detect in the image. If you know the length and the width of the picture, you can define one of its corners as the origin, and define the coordinates of the other three corners relative to it. Now you have four points in the world coordinates, and four corresponding image points.

By the way, please try using the Camera Calibrator App available in the Computer Vision System Toolbox. IMHO, it is much easier to use than Caltech. Also, there is a function called extrinsics, which computes the camera extrinsics given its intrinsics, a set of world points, and a set of corresponding image points.

[Omar]

Thank you so much for your support, Dima. The problem now is I couldn't evaluate the trial version of Computer Vision System Toolbox before I purchase it and start use it on a wider scale for all my experiments. It says that a representative will return to me from 1-3 business days which unfortunately never happened!

If a wall picture has to appear next to my object in my photos for calibration purposes using Caltech's toolbox, then it will not differ much than the checkerboard. In this case I can use any of them.

The new question that comes into the picture now if I used a checkerboard (or a photo on the wall), do you know by any chance how to take them out of the reconstructed 3D photo?

In other words, I want to return a 3D reconstruction to "only" the object I'm picturing without anything else.

Thanks once again.