Realistic modeling of spectator behavior for soccer videogames with CUDA

Soccer has always been one of the most popular videogame genres. When designing a soccer game, designers tend to focus on the game field and game play due to the limited computational resources, and thus the modelling of virtual spectators is paid less attention. In this study we present a novel approach to the modeling of spectator behavior, which treats each spectator as a unique individual. We also propose an independent software layer for sport-based games that simply obtains the game status from the game engine via a simple messaging protocol and computes the spectator behavior accordingly. The result is returned to the game engine, to be used in the animation and rendering of the spectators. Additionally, we offer a customizable spectator knowledge base with well structured XML to minimize coding efforts, while generating individualized behavior. The employed AI is based on fuzzy inference. In order to overcome additional demand for computing realistic spectator behavior, we use GPU parallel computing with CUDA.

Graphical AbstractFigure optionsDownload high-quality image (191 K)Download as PowerPoint slideHighlights
► We realistically model distinct behavior of crowd consisting of virtual spectators.
► We also implemented Mexican Wave in the stadium using the proposed approach.
► We used CUDA to process a crowd of virtual spectators on GPU in real-time.
► Computations required for simulation take very small portion of GPU processing time.