Dynamic and Scalable Large Scale Image Reconstruction
Recent approaches to reconstructing city-sized areas from large image collections usually process them all at once and only produce disconnected descriptions of image subsets, which typically correspond to major landmarks.
In contrast, we propose a framework that lets us take advantage of the available meta-data to build a single, consistent description from these potentially disconnected descriptions. Furthermore, this description can be incrementally updated and enriched as new images become available. We demonstrate the power of our approach by building large-scale reconstructions using images of Lausanne and Prague.
The calibration data has been used to learn a short binary descriptor in LDAHash: Improved matching with smaller descriptors .
The dense reconstruction of the scenes we calibrated here can be found on the Efficient Large Scale Multi-View Stereo for Ultra High Resolution Image Sets web page.
This work was supported in part by Nokia Research Center.
Results
Prague
Each calibrated image cluster is shown in a different color. The colored points correspond to the projection of the cluster 3D points onto the map. Green lines indicate the 2D building footprints which are available on openstreetmap.
|
position of the calibrated clusters using geo-tags only |
our final cluster alignment |
 |
 |
| click to enlarge | click to enlarge |
3D rendering of the Prague dataset.
Lausanne
Each calibrated image cluster is shown in a different color. The colored points correspond to the projection of the cluster 3D points onto the map. Green lines indicate the 2D building footprints which are available on openstreetmap.
|
position of the calibrated clusters using geo-tags only |
our final cluster alignment |
 |
|
| click to enlarge | click to enlarge |
3D rendering of the Lausanne dataset.
Berlin
3D rendering of the Berlin dataset.
References
Main Reference
Dynamic and Scalable Large Scale Image Reconstruction
2010. 23rd IEEE Conference on Computer Vision and Pattern Recognition, San Francisco, U.S.A, June 13-19, 2010. p. 406-413. DOI : 10.1109/CVPR.2010.5540184.Related References
LDAHash: Improved Matching with Smaller Descriptors
IEEE Transactions on Pattern Analysis and Machine Intelligence. 2012. Vol. 34, p. 66-78. DOI : 10.1109/TPAMI.2011.103.Efficient Large Scale Multi-View Stereo for Ultra High Resolution Image Sets
Machine Vision and Applications. 2012. Vol. 23, num. 5, p. 903-920. DOI : 10.1007/s00138-011-0346-8.Contacts
| Christoph Strecha | [e-mail] |
| Timo Pylvanainen | [e-mail] |
| Pascal Fua | [e-mail] |