DOI: 10.1111/cgf.13091 COMPUTER GRAPHICS forum Volume 00 (2017), number 00 pp. 1–14 Detail-Preserving Explicit Mesh Projection and Topology Matching for Particle-Based Fluids F. Dagenais 1 , J. Gagnon 1,2 and E. Paquette 1 1 Multimedia Lab, ´ Ecole de technologie sup´ erieure, Montreal, Canada francois.dagenais.2@ens.etsmtl.ca, eric.paquette@etsmtl.ca 2 Mokko Studio, Montreal, Canada jonathangagnon@gmail.com Abstract We propose a new explicit surface tracking approach for particle-based fluid simulations. Our goal is to advect and update a highly detailed surface, while only computing a coarse simulation. Current explicit surface methods lose surface details when projecting on the isosurface of an implicit function built from particles. Our approach uses a detail-preserving projection, based on a signed distance field, to prevent the divergence of the explicit surface without losing its initial details. Furthermore, we introduce a novel topology matching stage that corrects the topology of the explicit surface based on the topology of an implicit function. To that end, we introduce an optimization approach to update our explicit mesh signed distance field before remeshing. Our approach is successfully used to preserve the surface details of melting and highly viscous objects, and shown to be stable by handling complex cases involving multiple topological changes. Compared to the computation of a high-resolution simulation, using our approach with a coarse fluid simulation significantly reduces the computation time and improves the quality of the resulting surface. Keywords: fluid modelling, animation, physically based animation, animation ACM CCS: Categories and Subject Descriptors (according to ACM CCS): I.3.5 [Computer Graphics]: Computational Geometry and Object Modelling, Physically based modelling I.3.7 [Computer Graphics]: Three-Dimensional Graphics and Realism, Animation 1. Introduction Particles are commonly used in fluid simulation, particularly with the rise in popularity of the Fluid-Implicit-Particle (FLIP) method [ZB05] found in commercial software (such as Houdini TM , Maya R  , and Realflow TM ). When simulating liquids and reconstruct- ing the surface from the particles, the irregular distribution of parti- cles, as well as the spherical nature of the implicit functions typically used for reconstruction, make the resulting surface prone to bumpi- ness. This is shown in Figure 2, where the Stanford Armadillo is reconstructed using different numbers of particles. As can be seen, a coarser simulation will severely smooth the details; therefore, a large amount of particles is needed to accurately reconstruct the details of the original mesh. Even in Figure 2(d), where over four million particles were used, some details around the eyes and the teeth have been smoothed out. When melting objects using a highly viscous simulation, it is important to retain the details of the original mesh, and this requires a large number of particles. However, such a detailed simulation might be unnecessary, considering the small amount of detail present in the fluid movement. This makes the simulation and surface reconstruction times unnecessarily longer, whereas the objective is only to improve the reconstructed surface. Explicit surface tracking [WMFB11] helps solve this problem by using an initial mesh and evolving it based on the underlying simulation. Remeshing is done throughout the process in order to preserve the coherence and the quality of the surface. However, explicit surface tracking is not well suited for particle-based simu- lations since the explicit mesh tends to diverge from the particles. While recent work address this issue [YWTY12], they do not pre- serve surface details very well. Since they rely on a projection onto a surface constructed from the particles, the size of the explicit c  2017 The Authors Computer Graphics Forum c  2017 The Eurographics Association and John Wiley & Sons Ltd. Published by John Wiley & Sons Ltd. 1