Publication Details

GeoPose3K: Mountain Landscape Dataset for Camera Pose Estimation in Outdoor Environments

BREJCHA, J.; ČADÍK, M. GeoPose3K: Mountain Landscape Dataset for Camera Pose Estimation in Outdoor Environments. IMAGE AND VISION COMPUTING, 2017, vol. 2017, no. 1, p. 1-41. ISSN: 0262-8856.
Czech title
GeoPose3K: Dataset horské krajiny pro odhad pozice a orientace kamery v přírodním prostředí
Type
journal article
Language
English
Authors
Brejcha Jan, Ing., Ph.D. (RG CPHOTO)
Čadík Martin, doc. Ing., Ph.D. (DCGM)
URL
Keywords

camera pose estimation, visual geo-localization,cameraorientation estimation,image-to-model registration, digital terrainmodels, semantic segmentation 

Abstract

We introduce a new dataset called GeoPose3K which contains overthree thousand precise camera poses of mountain landscape images. Inaddition to camera location and orientation, we provide data for trainingand evaluation of computer vision methods and applications in context ofoutdoor scenes; synthetic depth maps, normal maps, illuminationsimulation and semantic labels. To illustrate properties of the dataset,we compare results achieved by state-of-the-art visual geo-localizationmethod on GeoPose3K with results achieved on an existing dataset forvisual geo-localization. To foster research of computer vision algorithmsfor outdoor environments, we propose several novel future use-cases ofour new GeoPose3K dataset. 

Published
2017
Pages
1–41
Journal
IMAGE AND VISION COMPUTING, vol. 2017, no. 1, ISSN 0262-8856
DOI
10.1016/j.imavis.2017.05.009
UT WoS
000413060000001
EID Scopus
2-s2.0-85028660327
BibTeX 
@article{BUT144445,
  author="Jan {Brejcha} and Martin {Čadík}",
  title="GeoPose3K: Mountain Landscape Dataset for Camera Pose Estimation in Outdoor Environments",
  journal="IMAGE AND VISION COMPUTING",
  year="2017",
  volume="2017",
  number="1",
  pages="1--41",
  doi="10.1016/j.imavis.2017.05.009",
  issn="0262-8856",
  url="https://www.sciencedirect.com/science/article/abs/pii/S0262885617300963?via%3Dihub"
}
Files
Back to top