Texture Map Generation for 3D Reconstructed Scenes

Junho Jeon
POSTECH

Yeongyu Jung
POSTECH

Haejoon Kim
POSTECH

Seungyong Lee
POSTECH

The Visual Computer (Special Issue on CGI 2016), Vol 32, Issue 6, June 2016


Framework Overview
 
Large Scene Examples

Abstract

We present a novel method for generating texture maps for 3D geometric models reconstructed using consumer RGB-D sensors. Our method generates a texture map for a simplified 3D mesh of the reconstructed scene using spatially and temporally sub-sampled key frames of the input RGB stream. We acquire an accurate texture map by optimizing the texture coordinates of the 3D model to maximize the photometric consistency among multiple key frames. We show that the optimization can be performed efficiently using GPU by exploiting the locality of texture coordinate manipulation. Experimental results demonstrate that our method can generate a texture map in a few tens of seconds for a large 3D model, such as a whole room.

Paper (preprint)
PDF(5.04MB)
Talk
PDF(3.49MB)
Supplementary Video
Youtube

BibTex

@article{Jeon2016,
author="Jeon, Junho and Jung, Yeongyu and Kim, Haejoon and Lee, Seungyong",
title="Texture map generation for 3D reconstructed scenes",
journal="The Visual Computer",
year="2016",
volume="32",
number="6",
pages="955--965",
}