Acoustic simulation based on multi-view images: A 3DGS-enhanced boundary element method

This study establishes a reconstruction-assisted acoustic simulation workflow for airborne acoustic scattering that links multi-view image reconstruction, geometry post-processing, Catmull–Clark subdivision surface representation, and BEM and FMM-BEM acoustic simulation. In this workflow, 3D Gaussian Splatting (3DGS) and SuGaR are used to obtain an initial explicit geometry from multi-view images. The reconstructed geometry is then post-processed to improve watertightness, normal consistency, and surface smoothness, and it is converted into a Catmull–Clark subdivision surface representation suitable for acoustic BEM. FMM is further used to reduce the cost of large-scale boundary element calculations. Numerical examples on a sphere, an engine model, and a vehicle model verify the underlying BEM formulation, examine reconstructed complex geometries, and compare conventional BEM with FMM-BEM. The results demonstrate the feasibility of converting multi-view reconstructed geometries into watertight and smooth models for acoustic scattering in air, reducing the dependence on manual geometry repair and providing a potential route for future acoustic simulation of measured engineering structures.