Abstract:
As we move towards improving the skill of computers to play games like chess against humans, the ability to accurately perceive real-world game boards and game states remains a challenge in many cases, hindering the development of game-playing robots. In this paper, we present a computer vision algorithm developed as part of a chess robot project that detects the chess board, squares, and piece positions in relatively unconstrained environments. Dynamically responding to lighting changes in the environment, accounting for perspective distortion, and using accurate detection methodologies results in a simple but robust algorithm that succeeds 100% of the time in standard environments, and 80% of the time in extreme environments with external lighting. The key contributions of this paper are a dynamic approach to the Hough line transform, and a hybrid edge and morphology-based approach for object/occupancy detection, that enable the development of a robot chess player that relies solely on the camera for sensory input.
