| Abstract [eng] |
Artificial Intelligence and Computer Vision Algorithms in Table Soccer Game Table soccer is a very fast paced strategic game, where there are two teams playing against each other and controlling their set of players. Ball that is in the game can reach speed up to 10 m/s which requires players to react very fast to ball position changes. This game is a very good field for artificial intelligence, computer vision and robotics research, because you have to pick a strategy to be able to score against your opponent, you have to be able to see the position of the ball and players and you have to be able to adjust your players as quickly as possible. Our main goal is to anayze computer vision and learning methods, create a working physical prototype of foosball table and create a program which is able to "understand" and follow red players, blue players, ball and make decision what to do using learning algorithm. In this paper we discussed related work, analyzed best tools and programming languages to use when creating computer vision algorithm and we decided to use Python with OpenCV, and HSV colour space instead of RGB. We analyzed the most important morphological methods (erosion, dilation, blur) in computer vision and explained how our created algorithm works. We conducted a thorough learning alghoritms analysis and decided to use Q-learning with action function approximation method to create our own learning alghoritm. We also constructed physical foosball table that is controlled with servo motors and we presented results that was achieved. For computer vision we used "Sony PlayStation 3 Eye" camera and discussed results - ball was found in 92.09% frames which is acceptable result. |
