Stand H10

About the Project

My project focuses on building a real-time chess detection system capable of translating real-world chess games into their digital equivalent. The first step in achieving this is board detection, which involves recognizing chessboard coordinates from a live webcam feed. Different computer vision strategies were applied, including a Hough transformation to find the intersection amongst the horizontal and vertical line segments to localize board coordinates. Accurate board detection is crucial because it establishes the coordinate system on which all later piece recognition depends. Once the board has been located, the next step is object detection for the chess pieces. This involves localizing each piece on the board and classifying it into one of the 12 categories (six white and six black pieces). Piece detection is achieved in in one of two paradigms, the first of which involves a two stage RCNN, defining regional proposals and classifying each. The other paradigm involves a single stage YOLO model which classifies and locates each chess piece in a single step. The final and most critical step is the combination of board and piece detection. By aligning the localized pieces with the detected board squares, the system generates a complete digital representation of the current position in Forsyth–Edwards Notation (FEN). FEN is a standardized way of describing chess positions, allowing seamless game digitization and interaction with chess engines. Once the FEN is produced, it can be fed into existing chess engines for tasks such as move validation, analysis, and strategy recommendations. This plug-and-play approach transforms a physical game into a digital one in real time, opening up possibilities for automated game recording, live broadcasting, and advanced player assistance.