Some comments on what you did. Using camera from your Laptop could be an issue since light could change and your OpenCV program will not be able to recognize the pieces. If you want to speed up on this task, I would advise you to take a software based board such as Winboard, Arena or the like. It will be easier.
What you propose here is already considered in some Github repos such as the really nice Deep Pink and Deep Murasaki and for Deep Learning positions, just google it and you will find several architectures
Finally regarding AlphaGo, check the Nature paper, and in general DeepMind’s website, there are different papers on the subject