| In this thesis, we describe a new method for implementing intelligent automated camera control in spectator games, specifically in replays of the video game Dota 2. First, we give a brief overview of Dota 2 and how spectator mode functions in games. We survey current camera control systems used in industry and conclude that current solutions use only simple heuristics to control the camera. We propose an improved solution that uses a hybrid of both machine learning and heuristics for performing camera control, which would be able to detect important and interesting events. Next, we describe an implementation of our solution in the form of a Python prototype. Finally, we present the results of testing our prototype on users found via the Amazon Mechanical Turk platform. We conclude that although our prototype did not fare well in testing, it could potentially replace current systems used in games. We also explore possible areas for future work, such as use in games other than Dota 2. |
