small_potato's profile - activity

When I use Camera Calibration Toolbox for Matlab, I found checkerboard placement has some impact on the variance of the measured parameters. Specifically, if the checkerboard is placed relatively far away from the camera, the uncertainty of both focal length and principal axis increase. The placement is far away enough that the checkerboard only occupies maybe a quarter of the image.

My questions are how the toolbox compute the variances, why placing the checkerboard far away can increase the variances.

Also, even the checkerboard is placed relatively close to the camera, there is still depth variation due to the orientation of the plane and some small travel of the plane in depth direction. If the focus does not change, I would assume that the calibration works for all depth besides the calibration depth, since the physical property of the camera has not changed at all. Is that a safe assumption?

Thank you in advance.

In the Matlab Calibration Toolbox, stereo image rectification is done via back projection:

ray = R' * inv(KK) * x

where KK is the camera intrinsic matrix and R represents the rotations required to rectify the image. [ray(1,:)/ray(3,:) ray(2,:)/ray(3,:)] represents pixels in the original image and x represents pixels in the rectified image.

In case I want to perform rectification in forward projection:

x = KK * R * x'.

[x(1,:)/x(3,:) x(2,:)/x(3,:)] represents the pixels in the rectified image. I am confused about how do I obtain the x', providing the locations of the pixels to be rectified in the original image.

I have a pair of matched 2D features extracted from rectified stereo image. Using cvPerspectiveTransform function in OpenCV, I attempted to reconstruct those features in 3D. The result is not consistent with the actual object dimension in real world. I realize there is a function in Matlab calibration toolbox that converts 2D stereo features into 3D point cloud. Nevertheless, the features are lifted from original images.

If I want to work with rectified images, is it possible to reconstruct the 3D locations based on 2D feature locations and disparity information.

If my target camera has a focus length of 1500 mm, what kind of depth range shall I place my calibration board? Is the depth range related to the calibration accuracy?

Thanks for the reply. It seems increasing the square size does help. Can you comment on how the square size is related to calibration accuracy?

I have been using Matlab Calibration Toolbox to calibrate my stereo camera. The result is not satisfying. I extract salient features from both images and verify the calibration result based on rectification result. After rectification, the features in the same pair are misaligned by roughly 10 pixels in y axis. I would expect the rectification error in y axis to be around 1 pixel.

Following the Opencv calibration example, for some images, the software fails to detect all the corner. Can someone comment on this? Also I have attached sample left and right images taken for calibration.

C:\fakepath\left0.jpg C:\fakepath\right0.jpg