CHESSBOARD SQUARE OCCUPANCY ANALYSIS THROUGH K-MEANS CLUSTERING BASED VISUAL MARKER DETECTION

Abstract

In 2012, Accessible chessboard for blind that is mouse and keyboard free was developed, using Kinect depth camera to detect presence of chess pieces. However, Kinect failed to be the right device for the task. Occasionally and unpredictably, depth data coming from the device contained large blobs of noise that reported zero distance from the Kinect. This prevented proper detection of square occupancy, requiring new algorithm that would be stable and robust. This paper presents such a new algorithm. It no longer users depth sensor, but USB web camera, and analyses captured image of the board. Colorful circular markers are stick to every square to help detection. Paper compares k-means clustering algorithm to other approaches that detect markers' visibility: Mathematical formulae for HSV statistics and neural network trained on the HSV statistics. Neural network based approach properly detected 19 out of 19 test images, but due to over fitting was failing in real game experiment. K-means clustering algorithm showed second-best results, detecting properly 16 out of 19 test images. Nevertheless it was chosen as the best algorithm due to its robustness. In an experiment of 3 chess games, all 15552 squares were detected properly by this winning algorithm.