automatic 2-D to 3-D geometric conversion

Carnegie Mellon researchers have developed software to automatically generate 3 D images from 2 D images. Watch the video and see how they even create the 3 D visualization from a painting.

The press release reports that the software imputes the figure geometry from the straight lines in the image and, I’m guessing from the documentation, also possesses some expert system qualities in understanding objects like ground, shadows, and clouds.

Using machine learning techniques, Robotics Institute researchers Alexei Efros and Martial Hebert, along with graduate student Derek Hoiem, have taught computers how to spot the visual cues that differentiate between vertical surfaces and horizontal surfaces in photographs of outdoor scenes. They’ve even developed a program that allows the computer to automatically generate 3-D reconstructions of scenes based on a single image.

the Carnegie Mellon researchers will show that having a sense of 3-D geometry helps computers identify objects, such as cars and pedestrians, in street scenes.

More detail, including the downloadable software, can be found here.

Now, if the software could embed geo-references at various points within the image then we could stitch together some high resolution walk-throughs from separate photos that also could be anchored to map locations. Think of the possibilities for flickr! I wonder what would be the minimum number of pairs of x,y coordinates that the photographers would have to submit as geo tags? We could render aerials as well but we’d have to orient the images to maximize the horizontal and vertical. Also, think of the possibility in taking old hand-drawn maps and street scenes. This could be be a wonderful addition to the work already being conducted on draping images onto digital elevation models.

Comments are closed.