In recent years, many computer models for pedestrian simulation have been developed to understand the performance of pedestrian evacuation under disasters. This study constructs a mesoscopic pedestrian simulation model to improve the efficiency of large-scale pedestrian evacuation simulation. This study also considers the dynamic environment, that is, the hazardous areas may vary over time and pedestrians can re-plan the path according to the updated environment. The developed mesoscopic model has two major features that improve simulation efficiency and maintain the accuracy of pedestrian simulation, which are multi-size cells and refined direction selection. Through the above features, this study drastically reduces the simulation time, enhances the accuracy of pedestrians' movement paths, and reflects the interactions between pedestrians and hazardous areas.