The Google Earth Blog posted about a presentation by Michael Jones from Google discussing the roadmap for Google Earth. It seems Google is working on analgorithm to automatically geocode uploaded photographs by comparing them with a large collection of known geo-tagged photos. These knowngeo-tagged photos could be drawn from a combination of Google Street View, Wikipedia, Picasa and other sources.
There also seems to be some work going on with landmark matching in particular — matching photos with landmarks (the Golden Gate Bridge as an example) as opposed to photos in general. Google published a research paper a few weeks ago that outlinesa system to match landmarks around the world with 80% accuracy. There is definitely some way to go before we see high accuracy for geocoding photos in general if this very specialized set of photos only manages 80% accuracy at present.
We’ll certainly be thinking about this technology and how it might relate to our digital asset management system,Sajara. In particular I wonder how their algorithms take into account how a neighborhood’s appearance can change over time. If an algorithm takes into account the date the photograph was taken this could be useful to automatically tag photos not only in regards to their location, but perhaps also the date they were taken.
By adding the power of geographic search and visualization to your collection, Sajara lets users retrieve and organize assets by:address, intersection, neighborhood, and geographic areas such as states and countries, along with more traditional search functions such as keywords, tags, time, and metadata.
The display of assets includes descriptive information, a map, thumbnail images, and video or audio streams, including the option to view assets in Google Maps or Google Earth.
Users can zoom in and out of the area, pan the map, making other assets appear on the map as they go, and enabling the search geographical parameters and thumbnails to update automatically.