Epipolar geometry estimated with a classical calibration technique

We know precisely the 3D coordinates of the corners of the checker pattern of the bi-plane calibration apparatus. The coordinates are measured in a reference frame attached to this apparatus, and they are available HERE. In total, there are 128 points. The corresponding points in the two images are automatically extracted, and their image coordinates are available HERE for the first image and THERE for the second image.

The camera perspective projection matrix can then be estimated for each camera by minimizing the distances between image points and the re-projections of their corresponding 3D points. The resulting files are available HERE for the first image and THERE for the second image.

From the two projection matrices, we can compute the fundamental matrix between the two cameras, which is available HERE.

What follows is Java applet which illustrates the estimated epipolar geometry. By clicking in one of the two images, the corresponding epipolar line of this point in the other image will be shown. (For better visualization, you'd better to enlarge the height of your browser such that both images are visible simultaneously.) You will find that the estimated epipolar geometry does not correspond to what we expected (vertical, parallel to each other).

If you see this text, then your browser might not support Java.



Author: Zhengyou Zhang
Email: zzhang@sophia.inria.fr
Last modified: Wed Oct 23 13:47:09 MET DST