Scalable Evacuation Simulation and Visualization Using GPU Computing

Abstract

In this paper, we will describe our novel approach to real-time visualiza- tion of crowdflowusing a scalable agent-based evacuation simulation that is suitable for execution on GPUs. To simulate the crowd behavior, we used the social force model in which each individual is represented by a self-driven particle subject to social and physical forces. The social force calculationswere executed using CUDA technology, a parallel-processing architecture for many-core GPUs. As a result, the GPU version was shown to have better scalability than the CPU version. In a case study of an evacuation scenario of a large underground shopping mall, the GPU version is approximately seven times faster than the CPU version and was capable of sustaining an interactive frame rate when visualizing the evacuation.