The new Google Earth 4.2 has a thoroughly-promoted planetarium feature, but quieter and less-noticed is the new Gigapan collaboration, whereby high-resolution photography is oriented to overlay the Google Earth representation from a specific angle, effectively allowing for a more detailed examination of that object while still retaining an idea of how those details correspond to the 3D-space. The video explains this better than I could in words, and there’s also a Data Mining blog post with decent-resolution pictures of the process.
This is a different - and less technically demanding - approach than Microsoft’s Photosynth technoology, which correlates hundreds of images to create a rough three-dimensional representation of an object, allowing you to view the object itself as a collection of points, with thousands of image-views to choose from, all automatedly oriented in the same way they were captured. Again, the process is easier to explain by trying it out on these Kennedy Space Center examples.
Both these examples are approximations of the real end goal, which is to not just create a 3D cloud of points, but a true 3D model of the object based off comparing all the pictures, with those pictures then used to project and map detail onto the rough 3D model. The effect is sort of like using an overhead projector to project and trace a larger version of a small drawing, except against something more complicated than a simple wall or whiteboard. This holy grail of image-data analysis would function something like the techniques Paul Debevec used to create his innovative Campanile Movie and, later, the background for that one shot in The Matrix. With that technology in place, it would be cheap, easy, and utterly effective to make a model of the Empire State Building (or any other landmark) by searching for that name on Flickr.
With more exacting versions of geotagging (associating pictures with the relevant latitude/longitude), the possibilities expand further. If GPS devices that recorded the position (and, possibly, the orientation) become embedded into the cameras we buy, it’ll become trivial to find all the pictures taken within a few hundred yards of a location. 3D modeling would suddenly become possible for not just the glamorous and widely-known locations, but also down to individual houses in neighborhoods. While some may cry foul to the potentially-invasive nature of such hyper-geotagging, Google’s already driving around with vans that capture the entire swath of area viewable from the street. And in any such case, the future has a habit of happening despite our strongest intentions otherwise.
Tags: Technology