Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An efficient GPU approach for designing 3D cultural heritage information systems

Luis López, Juan Carlos Torres, Germán Arroyo, Pedro Cano, Domingo Martín

TL;DR

The paper tackles the challenge of efficiently organizing and editing rich, heterogeneous data on 3D cultural heritage models. It proposes a GPU-driven architecture that stores information layers in 2D textures and indexes them with texture coordinates, avoiding octree indexing. The approach introduces two layer types (Numeric and Database) implemented via Data, Mask, and Palette textures, managed by a texture-array system and edited entirely on the GPU using TEA and TPA algorithms. Empirical results show the GPU-based method offers higher precision, lower CPU–GPU data transfer, and better scalability across model complexity and editing sizes compared to an octree-based baseline, making it well-suited for interactive, field-ready cultural heritage information systems.

Abstract

We propose a new architecture for 3D information systems that takes advantage of the inherent parallelism of the GPUs. This new solution structures information as thematic layers, allowing a level of detail independent of the resolution of the meshes. Previous proposals of layer based systems present issues, both in terms of performance and storage, due to the use of octrees to index information. In contrast, our approach employs two-dimensional textures, highly efficient in GPU, to store and index information. In this article we describe this architecture and detail the GPU algorithms required to edit these layers. Finally, we present a performance comparison of our approach against an octree based system.

An efficient GPU approach for designing 3D cultural heritage information systems

TL;DR

The paper tackles the challenge of efficiently organizing and editing rich, heterogeneous data on 3D cultural heritage models. It proposes a GPU-driven architecture that stores information layers in 2D textures and indexes them with texture coordinates, avoiding octree indexing. The approach introduces two layer types (Numeric and Database) implemented via Data, Mask, and Palette textures, managed by a texture-array system and edited entirely on the GPU using TEA and TPA algorithms. Empirical results show the GPU-based method offers higher precision, lower CPU–GPU data transfer, and better scalability across model complexity and editing sizes compared to an octree-based baseline, making it well-suited for interactive, field-ready cultural heritage information systems.

Abstract

We propose a new architecture for 3D information systems that takes advantage of the inherent parallelism of the GPUs. This new solution structures information as thematic layers, allowing a level of detail independent of the resolution of the meshes. Previous proposals of layer based systems present issues, both in terms of performance and storage, due to the use of octrees to index information. In contrast, our approach employs two-dimensional textures, highly efficient in GPU, to store and index information. In this article we describe this architecture and detail the GPU algorithms required to edit these layers. Finally, we present a performance comparison of our approach against an octree based system.
Paper Structure (17 sections, 2 equations, 8 figures, 1 table)

This paper contains 17 sections, 2 equations, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. Research Aim
  3. Previous works
  4. 2D space
  5. 2.5D space
  6. Hybrid space
  7. 3D space
  8. Annotation systems
  9. Layer based systems
  10. Proposed approach
  11. Texture as a heterogeneous data structure
  12. Information layer
  13. Editing layers on the 3D model
  14. Texture Editing Algorithm
  15. Texture Padding Algorithm
  16. ...and 2 more sections

Figures (8)

  • Figure 1: Prototype of the proposed architecture. The user has associated two information layers to the 3D model of a vessel. The first layer cracks identifies the area of the cracks using a green color. The second layer pieces identifies the different pieces of the shattered vessel using the selected palette.
  • Figure 2: General overview of the proposed architecture.
  • Figure 3: Our algorithm filters areas of the texture progressively. (a) Depth Test discards the occluded parts of the 3D model. (b) Projective Texturing projects a squared area of the appropriated size onto the remaining area. Finally, (c) Shape Masking discards the texels that fall outside of the editing tool shape.
  • Figure 4: In the zoomed area, the algorithm has verified that the current texel is part of the Outline Mask (black line) and the kernel (red matrix) is checking whether there are texels of the Edited Area Mask nearby (orange shape).
  • Figure 5: 3D models used in the tests. Models (a) and (c) have colors per vertex while (b) and (d) use a generic gray color. They are organized in order of geometric complexity from left to right. The dimensions are also detailed for each model.
  • ...and 3 more figures