Friday, May 20, 2011

#ALGORITHMS: "Cloud Makes 3D Models from Aerial Photos"

Cloud-based services are enabling fast, cheap, large-scale three-dimensional models of almost any landscape. The models are generated from easy-to-obtain aerial photos from drones—unmanned aerial vehicles.

Unmanned drones can take thousands of aerial photographs today, but stitching them together has required human expertise and sophisticated high-end software. (Source: EPFL)

New software from EPFL spinoff Pix4D automatically generates 3D models from aerial photos. (Source: EPFL)

Unmanned aerial drones (UAVs) are becoming inexpensive enough for small businesses or even individuals to use, permitting thousands of aerial photographs to be snapped of points of interest. Unfortunately, the high-powered analysis software required to stitch together aerial photos is outside the budget of all but large corporations. Now a new genre of inexpensive cloud-based services is appearing, capable not only of stitching together those patchworks of photos, but even able to automatically interpret what they see, thereby generating three-dimensional (3D) models on the cheap.

The Pix4D project does just that. A spin-off of the European research organization called the Ecole Polytechnique Federale de Lausanne (EPFL), Pix4D was named for its ability to transform the fourth dimension—time—into a method of generating 3D models from 2D images shot by aerial drones. By harnessing time, a UAV with a digital camera can take thousands of photographs from the air, capturing every possible angle of view of objects on the ground. Without smart cloud-based computing resources, however, these aerial photos would have to be hand-assembled, and even then they would only yield a flat 2D map of the area photographed.

Pix4D software running in the clouds, on the other hand, not only automatically stitches together thousands of 2D images to make accurate maps, but can also infer the 3D information needed to make a model that can then be viewed from any orientation. The cloud service works with the geo-tags on each image, comparing them with those taken at nearby times and locations, resulting in a stunning 3D model of whatever is imaged using a relatively inexpensive cloud-based service.

The Pix4D cloud service accepts a stream of related photos from which it generates a 3D model in as little as 30 minutes. The service not only automatically generates the 3D maps, but also adds points of interest that can be cataloged by users. To demonstrate the service, Pix4D took 50,000 photos of its host city—Lausanne, Switzerland—and created the world's highest-resolution 3D model of the city. The Pix4D user interface then allows users to navigate to any location in the city and view it from any orientation.

New software from EPFL spinoff Pix4D automatically generates 3D models from aerial photos. (Source: EPFL)
Don't have a ready UAV? EPFL has spun off another startup that makes an inexpensive drone. Called the senseFly, this pint-sized aerial vehicle is currently being used to take high-resolution photos for many applications, from farmers who wish to survey the evolution of their crops over large distances and long periods of time to archaeologists hunting for evidence of as yet undiscovered ruins.
Further Reading: