I have a project and it takes a picture of a screen to find a laser dot’s location. On the corners of the screen, there will be 4 chessboards. The camera could potentially be bumped into and the angle can be different.
I will be taking and saving a reference image to find the chessboard corners of all four chessboards. I will know know that this is where those chessboards should be. Then, I calibrate the camera by taking another image and it will do a similar thing and finding the chessboard corners again. If the camera’s angle or position was changed, then those pixel locations should be different. With the homography function and perspective warp, I can make the test image in the calibration phase essentially look like the reference image. I know there are limits to this due to the FOV being out of the “linear” region of the camera. That’s why I also have the same thing as online check deposits on a banking app where you have to fit the screen in a square to know that the calibration with chessboard corners will work.
Maybe there is an entirely different approach to this I should take. I do not know what I should do though, so this lead me to keep trying to do this with the chessboard corners.
I have been struggling immensely with finding the chessboard corners. Since I have 4 chessboards, I crop the image around the chessboards individually with the help of Sobel edge detection. Also, I been using a multitude of techniques such as CLAHE, adaptive filtering, Gaussian blurring, and color segmentation.
I think the main issue is how small the chessboards are in the entirety of the image. It makes them have less resolution and become grainy, which results in the corners of the chessboards to not seem as connect, which is why I think this is not working. I am kind of bottle necked, the screen is 3’ x 3’ and the plates in the corners that would hold the chessboards are 8" x 6." Also, I am not sure if I should use a 4x3 chessboards which has less points and makes the actual squares bigger, or a higher one like 9x6.
Any help would be greatly appreciated.