The goal of our project was to build an automatic tracking system for identifying a chess game. Our system should identify the move in the game
Abstract
The goal of our project was to build an automatic tracking system for identifying a chess game. Our system should identify the move in the game (the frame in which the move was completed) and after the move is identified, the system starts to “scan” all the checkers in the board and
identify their content. In this method we can identify a the moves in the game. The system is receiving a video movie of Chess game, and knows what moves have been done. This system is canceling the need in human observer in order to track the game.
Main problems and solutions:
Problem: we encountered problems with the camera that made our checker pictures to be in poor resolution,
very noisy and hard to resolve between the soldier and the background (the colors look alike) Solution: every checker is passed through a filter that cleans the background of the checker and parts that are known that they don’t contain a soldier pixel are hidden with a mask.
Problem: we are identifying the soldiers with a SVM classifier. the SVM classifier is very sensitive to shifting and rotation.
Solution : in every checker we are identifying the center of the soldier and placing it in the center of the checker,in addition we are syntheticaly multiplying our data base training set by rotating the checker.
Example : Image before preprocessing
Example : Image after processing
Conclusions
We are identifying the checkers in ~90%efficiency.
In order for receiving better results
we need to have better quality of th pictures.
Example
The board we had to identify:
Figure 1 : picture of the board we had to identify
Figure 2 : The tools we identify
Acknowledgment
We are grateful to our project supervisor Ehud Orian for his help and guidance throughout this work.
We are also grateful to the Ollendorff Minerva Center Fund for supporting this project.