A comparison of classification algorithms for chess pieces detection

This work presents an evaluation of different classification algorithms to be implemented in a computer vision system to detect chess pieces. The implementation consists of a mechatronic system that allows a person to play chess against a robot manipulator and on a standard board. The system is based on an industrial type robot, a webcam-based artificial vision subsystem and open software for the game engine. The vision subsystem features a webcam that captures photos from the board. The aim is to detect, by means of classification techniques, whether a square is occupied by a piece, and in such a case whether the piece is black or white. Using matrix models, it is possible to determine the last movement executed by the human opponent. Robotics applied to interactive games is an excellent problem to explore human-robot collaboration as it presents a structure whose complexity can be gradually increased.