Real-time Snow Simulator using Iterative-relaxation and Boundary Handling

Abstract: Background Physics-based snow simulation in real time is an unexplored area, the reason being the difficulty introduced by the multitude of factors that affect the snow behaviour, such as cohesion, thermodynamics, and compression. Simulating snow in real time when considering these factors can become computationally demanding. However, the continued advancement of graphics processing units makes the exploration of real-time snow simulation attractive. Recently published research on real time physics-based snow simulation shows promising results in a parallel solution and will serve as motivation and base for this thesis. Objectives This thesis aims to improve the time-step of a previously proposed method using an iterative method and improve the snow behaviour with a particle-based boundary handling implementation. The aim consists of the following objectives. Integrate an iterative method, extend the snow behaviour with additional snow types, and implement a particle-based boundary handling method with two-way coupling. The proposed method should remain comparable to the original method in terms of snow behaviour. In order to gather results, the methods are measured in performance and used in a questionnaire to analyse the behaviour. Methods An iterative method along with a particle-based boundary handling method is implemented. The methods are both measured and compared using quantitative tests. Additionally, a questionnaire is deployed to gather qualitative results about the behaviour of the snow. Results The proposed method outperforms the original method in terms of time-step size. The proposed method is capable of increasing the time-step tenfold while decreasing the execution time by approximately eight times. Finally, the results from the questionnaire verify the perceived naturalism of the snow and its comparability to the original method. Conclusions The proposed method can perform with an increased time-step and a lower execution time compared to the original method, at the cost of time spent per frame. Lastly, the snow is perceived as natural with the boundary handling method at a significance level of 1 %.