Skip Navigation eScience Home | ANU Home | Search FEIT | Search ANU

Faculty of Engineering and Information Technology (FEIT)

Dept. of Computer Science (DCS)

eScience < Master or eScience Diploma or Guest Student or DCS> Project

---

Title : Full Geotagging of Photographs : an googlemap mashup // Use of fully geotagged Images : projection in a 3D world // Voronoï Diagram on a sphere for an advanced 3D billboard

This project is designed to be easily divided in two or 2 or 3 independent parts for individual students.

• Status : not finalized proposition OR proposition OR booked by OR in progress by OR done by

• Student : <name and e-mail of the student>

• Tutor/Client : Pascal Vuylsteker <pvk@vuylsteker.net>

• Period : S1 2006

• Keywords : Geotagging Geolocation Photographs EXIF Billboard Voronoï Diagram

• Report (link to the final report)

---

Description

With the explosion of social projects like the photo sharing web site flickr [Flickr 2005], we get a more and more vast amount of pictures available to everybody and more and more often those images are geotagged [GeoBlogger 2005], meaning that a Longitude/Latitude information is provided with the image, either in the EXIF data or as a tag from the hosting site. Most of the tools exploiting these metadata will allow you to see thumbnail of the pictures, tacked up to a 2D map, usually in the form of a 2D web2.0 Mashup [Mashup 2005] of Google Map [GeoBlogger, Pixagogo, Frappr, SmugSmug , Panoramio 2005]. On should notice that it is not clear in those services if you are expected to give the location of the object in the photo or where the camera was located.
In that regards, the relatively most advanced projects [FlickrFly 2005] add some viewpoint type of information in the form of the following additional tags : ge:range (range of viewpoint in metres from the geotagged coordinates), ge:heading (angle of coordinate from the viewpoint in degrees where 0 is looking North) and ge:tilt (angle of view in degrees where 0 is straight down and 90 is from ground level).
This is just some additional information to specify the path of the fly done trough google earth from a contextual point towards the location of the camera. One could expect that the “heading” target represent the orientation of the camera, but it is not required, as for the tilt, it will rarely match the pitch of the camera as it will more appropriately be used to elevate the entry viewpoint at a high level to show the geographic context of the shoot.
From those observations, This project will explore a middle path between full computer vision and simple map pining of pictures. By adding some complete information about the photographs (by mainly adding the 3D orientation of the camera), and by using the focal length information, we should be able to position sets of pictures the way it was done in the “Invisible Shape of Things Past” project.

Outcome

The additional step would be to work on the clustering of the pictures. Statistically, it is probable than digital tourists will take pictures following two typical configurations either all in the direction of a central object (Eiffel Tower, Sydney Opera…) or from a usually small platform:.the “lookout”.
We just have to look for clusters of converging or diverging vector among the ray defined by the location and orientation of the camera to identify such configurations.
Once isolated one can imagine to project the set of images outside or inside (according to the type of configuration) a sphere centred on the point of convergence of the lines to produce an approximation of a panoramic image with overlapping images or in the case of converging set of pictures, a QTVR object like representation.

Project Part one : data acquisition and geotagging : a googlemap mashup

Using the Google Map API, the project should allow anybody to upload a picture to a web site (or select a picture from flickr, or select a picture) and to use convenient tools to specify not only the position of the camera (already been done elsehere !) but its orientation too, by clicking on a google map to match the camera position (!) and different points on the image (landmarks, horizon ...). From those additional point, You are then able to extract the orientation of the camera and or the location of the main object on the shoot. The produced information would either be written back into the EXIF field of the jpeg image or as tag on flickr. By uploading a GPS track, and by using time synchronisation, you don't even have to pinpoint the position of the camera : you already get a good 3D location.

Project Part two : using the information to build a 3D billboard in OpenGL

Once you have fully geotagged images, you are able to display them precisely into a 3D world. The information you use are : the focal length of the lens, the position/altitude of the camera and its direction. You still have to decide where on such a defined ray you want to project a texture of the picture. The focal length could be a indicator. It will definitely be used to define the scale of the texture. You may use the position of the main object if it had been provided. Chance are that the position of the main object won't be available. But once you get a statically significant number of pictures taken around a given area, you should be able to cluster those pictures into 2 categories : the ones all looking towards the same object and the ones all shoot from the same area. In both case, this cluster will define a point, centre of a sphere on which to project the pictures (outside or inside of the sphere depending on the type of cluster.

Project Part three : optimisation of the display using Voronoï Diagram on a sphere

If you look a the previously defined sphere, you may want to reduce the size of that sphere and display a limited number of image, according to your viewpoint in the 3D world. One way to decide which picture to display is by using the voronoï diagram of the projections of the camera positions on the sphere and by using the "closest" point in the voronoï diagram to the projection of the Viewpoint

More Information

- [Art+Com 1995] Invisible Shape of Things Past (1995) : http://www.artcom.de/index.php?option=com_acprojects&page=6&id=26&Itemid=144&details=0&lang=en
- [Community pano] A panorama oriented communityWalk : http://www.communitywalk.com/map/110
- [Flickr 2005] photo sharing web site : Flickr : http://www.flickr.com/
- [FlickrFly 2005] Flickr-Fly : One click in a photos on flickr launch google earth to fly towards the location of the picture : http://www.roblog.com/flickrfly-docs/
- [Frappr 2005] Create your Map http://www.frappr.com/
- [GeoBlogger 2005] Set of tool to to visualise interact with flickr geotagged photographs : http://www.geobloggers.com/
- [Mashup 2005] Web2.0 = (Ajax + open API )2 => When new web services are built by gluing together existing one : http://www.programmableweb.com/
- [Pixagogo 2005] Create your Photo map : http://maps.pixagogo.com/
- [CommunityWalk 2005] Link any digital data to map : http://www.communitywalk.com/
- [Panoramio 2005] Still another photo / mapping web site, well integrated : http://www.panoramio.com/
- [Ping 2005] Ping Planet Earth : open source GIS : http://www.planet-earth.org/
- [SmugSmug 2005] Like Filckr, but with integrated maps : http://www.smugmug.com/
- [Syren 2004] Daniel Woo : Syren – A Ship Based Location-Aware Audio Experience : http://www.gmat.unsw.edu.au/gnss2004unsw/WOO,%20Daniel%20P210.pdf
- [Zheng 2003] J. Y. Zheng, Digital Route Panorama, IEEE Multimedia, Vol. 10, No. 3, (July-Sept. 2003) : http://www.cs.iupui.edu/~jzheng/RP/

Tips / details

The development have to be either multiplateform (Java OR C + multiplateforme GUI  OR Web2.0/AJAX...) OR MacOSX

(not exclusive 'OR')