Complex Event Detection and Event Based Retrieval in eSports (Master Thesis, Finished)
In recent years, the competitive playing of video games by both professionals and amateurs, known as eSports, has experienced staggering growth. As players become more numerous and professional competition becomes tougher, innovative approaches are needed to help players and coaches analyze the games they played and provide insightful feedback. In this thesis, we present a novel way of high-level event detection and analysis in the video game DotA 2. The system we built handles the tasks of obtaining meaningful data from DotA replay files, complex event detection, and web-based retrieval. We will use StreamTeam, a modular, workflow-based stream processing system to implement a workflow capable of detecting high-level events, such as team fights and sieges, and provide the user with an interactive way of displaying these events using SportSense, a sketch-based video and event retrieval system. We will demonstrate the utility of our system in a user study, and evaluate the quality of our event detection against ground truth data.
Start / End Dates
2018/09/24 - 2019/03/23