Radiometry propagation to large 3D point clouds from sparsely sampled ground truth

Thomas Höll, Axel Pinz

Research output: Chapter in Book/Report/Conference proceedingConference paperpeer-review

Abstract

Good radiometry of a 3D reconstruction is essential for digital conservation and versatile visualization of cultural heritage artifacts and sites. For large sites, "true" radiometry for the complete 3D point cloud is very expensive to obtain. We present a method that is capable to reconstruct the radiometric surface properties of an entire scene despite the fact that we only have access to the "true" radiometry of a small part of it. This is done in a two stage process: First, we transfer the radiometry to spatially corresponding parts of the scene, and second, we propagate these values to the entire scene using affinity information. We apply our method to 3D point clouds and 2D images, and show excellent quantitative and visually pleasing qualitative results. This approach can be of high value in many applications where users want to improve phototextured models towards high-quality yet affordable radiometry.
Original languageEnglish
Title of host publicationComputer Vision – ACCV 2016, Part II
Subtitle of host publication13th Asian Conference on Computer Vision
EditorsC.S. Chen, J. Lu, KK. Ma
Place of PublicationCham
PublisherSpringer
Pages222-235
ISBN (Electronic)978-3-319-54190-7
ISBN (Print)978-3-319-54189-1
DOIs
Publication statusPublished - 2017
Event13th Asian Conference on Computer Vision: ACCV 2016 - Taipei International Convention Center, Taipei, Taiwan, Province of China
Duration: 20 Nov 201624 Nov 2016
Conference number: 13

Publication series

NameLecture Notes in Computer Science
Volume10117

Conference

Conference13th Asian Conference on Computer Vision
Abbreviated titleACCV
Country/TerritoryTaiwan, Province of China
CityTaipei
Period20/11/1624/11/16

Fingerprint

Dive into the research topics of 'Radiometry propagation to large 3D point clouds from sparsely sampled ground truth'. Together they form a unique fingerprint.

Cite this