Discretization and OpenCL-based implementation of social field pedestrian model

Abstract

Abstract A discretization is applied to the continuous field-based pedestrian model to derive a discrete version. The concept of walk period is introduced so that pedestrians’ varied walking velocities can be studied without introducing any modification to the underlying logic. This further allows the discrete model to analyze a flow of randomly mixed population. Next, starting from the discrete model, a high performance GPU model targeting at the OpenCL heterogeneous parallel framework is developed. Advanced techniques are developed to allow efficient access to stored field strength, avoid atomic operations, and so on. Numerical experiments disclose that the GPU model can bring a as high as 30.8x performance improvement, with contrast to the previously developed CPU model. Thus, when equipped with proper functional modules such as a macroscopic/mesoscopic route selection and so on, the GPU model can act as a core engine for simulation of a large/very-large population.