The OpenSFM Database

Matthias Adorjan
The OpenSFM Database
[thesis]

Information

Abstract

Besides using high-cost laser scanning equipment to capture large point clouds for topographical or architectural purposes, nowadays other, more affordable approaches exist. Structure-from-motion (SfM) in combination with multi-view stereo (MVS) is such a low-cost photogrammetric method used to generate large point datasets. It refers to the process of estimating three-dimensional structures out of two-dimensional image sequences. These sequences can even be captured with conventional consumer-grade digital cameras.

In our work we aim to a establish a free and fully accessible structure-from-motion system, based on the idea of collaborative projects like OpenStreetMap. Our client-server system, called OpenSfM, consists of a web front-end which lets the user explore, upload and edit SfM-datasets and a back-end that answers client requests and processes the uploaded data and stores it in a database.

The front-end is a virtual tourism client which allows the exploration of georeferenced point clouds together with their corresponding SfM-data like camera parameters and photos. The information is rendered in the context of an interactive virtual globe. An upload functionality makes it possible to integrate new SfM-datasets into the system and improve or extend existing datasets by adding images that fill missing areas of the affected point cloud. Furthermore, an edit mode allows the correction of georeferencing or reconstruction errors.

On the other side the back-end evaluates the uploaded information and generates georeferenced point datasets using a state-of-the-art SfM engine and the GPS data stored in the uploaded images. The generated point clouds are preprocessed, such that they can be used by the front-end’s point cloud renderer. After that, they are stored together with the uploaded images and SfM parameters in the underlying database.

On the whole, our system allows the gathering of SfM-datasets that represent different sights or landmarks, but also just locally famous buildings, placed all over the world. Those datasets can be explored in an interactive way by every user who accesses the virtual tourism client using a web browser.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@mastersthesis{Adorjan-2015,
  title =      "The OpenSFM Database",
  author =     "Matthias Adorjan",
  year =       "2015",
  abstract =   "Besides using high-cost laser scanning equipment to capture
               large point clouds for topographical or architectural
               purposes, nowadays other, more affordable approaches exist.
               Structure-from-motion (SfM) in combination with multi-view
               stereo (MVS) is such a low-cost photogrammetric method used
               to generate large point datasets. It refers to the process
               of estimating three-dimensional structures out of
               two-dimensional image sequences. These sequences can even be
               captured with conventional consumer-grade digital cameras. 
               In our work we aim to a establish a free and fully
               accessible structure-from-motion system, based on the idea
               of collaborative projects like OpenStreetMap. Our
               client-server system, called OpenSfM, consists of a web
               front-end which lets the user explore, upload and edit
               SfM-datasets and a back-end that answers client requests and
               processes the uploaded data and stores it in a database. 
               The front-end is a virtual tourism client which allows the
               exploration of georeferenced point clouds together with
               their corresponding SfM-data like camera parameters and
               photos. The information is rendered in the context of an
               interactive virtual globe. An upload functionality makes it
               possible to integrate new SfM-datasets into the system and
               improve or extend existing datasets by adding images that
               fill missing areas of the affected point cloud. Furthermore,
               an edit mode allows the correction of georeferencing or
               reconstruction errors.  On the other side the back-end
               evaluates the uploaded information and generates
               georeferenced point datasets using a state-of-the-art SfM
               engine and the GPS data stored in the uploaded images. The
               generated point clouds are preprocessed, such that they can
               be used by the front-end’s point cloud renderer. After
               that, they are stored together with the uploaded images and
               SfM parameters in the underlying database.  On the whole,
               our system allows the gathering of SfM-datasets that
               represent different sights or landmarks, but also just
               locally famous buildings, placed all over the world. Those
               datasets can be explored in an interactive way by every user
               who accesses the virtual tourism client using a web browser.",
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2015/Adorjan-2015/",
}