Abstract: In order to simulate small-scale geometrical features of Lagrangian particle fluid surface, this
paper presents an efficient octree-based adaptive fluid surface reconstruction approach. First, this
approach implements the more accurate detection of fluid surface particles. Then, the approach
adaptively decomposes the fluid particles’ AABB bounding boxes including fluid surface particles
based on octree. Finally, the implicit distance field is only constructed in the leaf nodes including fluid
surface particles. The memory consumption and computation complexity of fluid surface reconstruction
depend on fluid surface not fluid volume, so our method is very suitable for large scale fluid scenes.

Key words: Lagrangian particle, surface reconstruction, fluid simulation, realistic rendering