• Get Luxonis DepthAI library from https://github.com/luxonis/depthai-core
• Build it according to instructions on their repository's readme
• Build G-API with OpenCV AI Kit support:

cmake /path-to-opencv -DWITH_OAK=ON -Ddepthai_DIR=/path/to/depthai/library

Note: currently OpenCV AI backend is rather limited in terms of covering DepthAI's functionality. See G-API's samples for usage examples (e.g. OpenCV AI Kit inference and Getting video from OpenCV AI Kit device).