Camera calibration toolbox and Matlab view

Hi, i'm building a model to simulate camera calibration, image acquisition and reconstruction of a point in 3D space. For the first step i'm using Camera Calibration Toolbox from Caltech but i'm having some troubles: i've created a 3D view of a typical pattern (chess board) located in the Z-X plane with the bottom left corner placed in the origin.

In the file attached you can see a view of the pattern with camera placed at [20 20 25] and view angle 50, every square is 3x3.

Running the toolbox on one ore more views of the pattern from different angles i'm getting always extremely high focal lenghts (like 50.000 or even more) and warning about matrix singularity. It's not really clear to me how the figure view in Matlab works because if i use imshow and then get(gca) the camera position i get is [401 300.5 5004.5] (the centre is correct since the images i'm using are 800x600 but i don't understand the third coordinate)

Where am i making a mistake? How can i get Matlab calibration matrix to go back to 3D coordinates from pixel location of a point in 2 different images?

I need to simulate the calibration process to estimate calibration matrix and apply correct image rectification, so depth of the point in space will be found by triangulation or re-projecting the points of the images back in 3D space.

Thank you for your attention

[Chuanhao Li]

Hi, can you tell me how to create a 3D chessboard in Matlab? I am working on a project which requires me to generate some ground truth images for camera calibration. And I have worked out a model to project 3D points to 2D plane through the projection matrix I have set. But I don't know how to create a chessboard.

Answer 1

[Dima Lisin]

If you are still interested in calibrating cameras, please try the Camera Calibrator app in the Computer Vision System Toolbox.