jimfred's profile - activity

What capabilities does OpenCV have for a 'Mesh Warp' pixel remap where pixels of 2D image can be remapped to different locations based on an array of mapping control points? I've seen and used the remap function but if I understand it correctly, its map matrix specifies the location of every pixel. As an example, Id like to remap an image based on, say, a 3x4 matrix of mapping control points and all pixels in between would be remapped based on an interpolation algorithm (linear or preferably a higher order interpolation algorithm). I don't need to remap colors or intensity - just pixel locations.

The 'Camera calibration With OpenCV' pages says: 'you will probably need at least 10 good snapshots of the input pattern in different positions'. I think that explains why a single sample did not result in good calibration.

I followed the example in this tutorial: http://www.aishack.in/2010/07/calibrating-undistorting-with-opencv-in-c-oh-yeah/ ...and I got these results... ...where the middle image shows successful 3x7 corner detection but the corrected image on the right doesn't 'flatten' the image as expected.

One deviation from the tutorial: I only sampled the checker board once instead of multiple time. The image sampled is the image on the left. A post I found online suggested that too-few samples would result in calibration problems. I can, of course, post my values in the matrices updated by the calibrateCamera() call - I wanted to start with a short post.

Will a single sample work with calibrateCamera()?

A follow-on question: Could I do a similar remap using remap() with just a 3x7 matrix of points? The examples I found that use remap() appear to specify a matrix for ALL pixels to be remapped. I'd like to try a remap by specifying where I want the 3x7 matrix of points to appear. This would allow me to skip the calibrateCamera() call and use a more arbitrary remapping.