Real time simulation of virtual human crowd planning

Abstract

Behavioral simulation consists to simulate and animate virtual environments populated by virtual humans, and focuses both on local and global realism. For this reason, the simulation of pedestrians crowd is widely used by several domains such as the film industry, video games, security, civil engineering, urban planning ... etc. To simulate realistic crowds of virtual humans in real time, three main requirements need satisfaction. First of all, efficiency is essential, the simulation model must make it possible to simulate crowds in an efficient way in terms of computational cost. Secondly, quantity, that is to say, the ability to simulate thousands of characters, finally realism, which means the need to identify the interactions between each individual and his neighborhood and then influence the individual behaviors, and to reproduce some macroscopic phenomena. Proposing a solution able to manage all these three aspects is a challenging problem that we have addressed in this thesis. In this thesis, we develop a hybrid architecture to perform crowd behavior simulation in order to simultaneously satisfy the criterion of macroscopic and microscopic realism, while ensuring path planning and dynamic avoidance of collisions for large numbers of pedestrians. This architecture consists to divide the simulation environment into exhaustive regions, where motion modeling is managed by two approaches of different levels of detail. Concretely, the high-density regions are governed by a macroscopic approach based on a flow potential to generate macroscopic phenomena, while the other zones use a microscopic method to perform individual behaviors. Our architecture also allows to ensure the continuity of movements during a change between two algorithms.