A REAL-TIME EFFICIENT PARTICLE FILTER ALGORITHM FOR REALISTIC TARGET TRACKING IN GAMES

The challenge for a computer game agent in a partially observable targettracking environment is to estimate where the target might be before taking the next action. One technique that can be used for estimating the state of a hidden target is known as particle filtering. This project examines several variations of a widely-used particle filter algorithm. The project proposes modifications that reduce the algorithm’s computational complexity making it suitable for game applications with rigorous real-time requirements. As part of this project, both a graphical, and a command-line-based demonstration game are implemented to serve as the test environment. The project empirically examines the proposed unweighted particle filter algorithm. Evidence shows, using the algorithm, an agent in pursuit of a target exhibits realistic behavior and a significantly high skill level.