Hybrid Scene Structuring for Accelerating 3D Radiative Transfer Simulations - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Remote Sensing Année : 2019

Hybrid Scene Structuring for Accelerating 3D Radiative Transfer Simulations

Résumé

Three-dimensional (3D) radiative transfer models are the most accurate remote sensing models. However, presently the application of 3D models to heterogeneous Earth scenes is a computationally intensive task. A common approach to reduce computation time is abstracting the landscape elements into simpler geometries (e.g., ellipsoid), which, however, may introduce biases. Here, a hybrid scene structuring approach is proposed to accelerate the radiative transfer simulations while keeping the scene as realistic as possible. In a first step, a 3D description of the Earth landscape with equal-sized voxels is optimized to keep only non-empty voxels (i.e., voxels that contain triangles) and managed using a bounding volume hierarchy (BVH). For any voxel that contains triangles, within-voxel BVHs are created to accelerate the ray–triangle intersection tests. The hybrid scheme is implemented in the Discrete Anisotropic Radiative Transfer (DART) model by integrating the Embree ray-tracing kernels developed at Intel. In this paper, the performance of the hybrid algorithm is compared with the original uniform grid approach implemented in DART for a 3D city scene and a forest scene. Results show that the removal of empty voxels can accelerate urban simulation by 1.4×~3.7×, and that the within-voxel BVH can accelerate forest simulations by up to 258.5×.
Fichier principal
Vignette du fichier
remotesensing-11-02637.pdf (3.63 Mo) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte

Dates et versions

hal-04644165 , version 1 (11-07-2024)

Licence

Identifiants

Citer

Jianbo Qi, Tiangang Yin, Donghui Xie, Jean-Philippe Gastellu-Etchegorry. Hybrid Scene Structuring for Accelerating 3D Radiative Transfer Simulations. Remote Sensing, 2019, 11 (22), pp.2637. ⟨10.3390/rs11222637⟩. ⟨hal-04644165⟩
52 Consultations
5 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More