Combining video and sequential statistical relational techniques to monitor card games
Laura Antanas, Bernd Gutmann, Ingo Thon, Kristian Kersting, Luc De Raedt · Lirias (KU Leuven) · 2010
Games are a multi-billion dollar industry and a driving force behind technology. The key to make computer games more interesting is to create intelligent artificial game agents. A first step is teaching them the protocols to play a game. To the best of our knowledge, most systems which train AI agents are used in virtual environments. In this work we train a computer system in a real-world environment by video streams. First, we demonstrate a way to bridge the gap between low-level video data and high-level symbolic data. Second, using the high-level, yet noisy data, we show that state-of-the-art statistical relational learning systems are able to capture underlying concepts in video streams. We evaluate the selected methods on the task of detecting fraudulent behavior in card games.