The computer modelling of fallen snow

• Main Author: Fearing, Paul
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/11075

One of nature's greatest beauties is the way fresh snow covers the world in a perfect blanket of crystalline white. Snow replaces sharp angles with gentle curves, and clings to surfaces to form ghostly silhouettes. It is said the Inuit have 50 different words for snow, yet even they can be left speechless, as snow is one of the most complex natural materials in existence. This research presents a new model of visual snow accumulation for computer graphics. We are primarily concerned with creating and simulating fallen snow (not falling snow - an important distinction), with our ultimate goal to produce view-independent, static, 3D snow surface models that can be used in artistic and scientific visualisation, film, and advertising. Our contribution is divided into two major components: snow placement and snow stability. Each are essential for modelling the appearance of a thick layer of snowfall on the ground. Snow placement requires us to determine how much snow falls upon the scene, and where it accumulates. We simulate this with an adaptive particle/surface hybrid system that allows for such phenomena as flake flutter, flake dusting and wind-blown snow. We compute snow accumulation by shooting particles upwards towards the sky, giving each source surface independent control over its own sampling density, accuracy and computation time. Importance ordering minimises sampling effort while maximising visual information, generating smoothly improving global results that can be interrupted at any point. Once snow lands on the ground, our stability model moves material away from physically unstable areas in a series of small, simultaneous avalanches. We use a simple local stability test that handles very steep surfaces, obstacles, edges, and snow transit due to wind. Our stability algorithm is flexible enough to simulate other materials, such as flour, sand, and flowing water. We show physical plausibility by comparing various aspects of our approach with real snow images. As proof that our algorithm is flexible and usable, we provide several examples of snow on complex models containing hundreds of thousands of polygons. The completed 3D snow surface model can be easily imported into commercial modelling and rendering software, allowing users to convert existing animations to a brand new season.