WANT a detailed three-dimensional model of your house or back garden? Just stick up a webcam and wait. Current 3D models of outdoor places, such as those seen in Google Earth, are typically created using laser scans or painstaking manual measurements. But anyone may soon be able to make one – using just daylight, time and a humble webcam.
It is all down to how sunlight bounces off surfaces differently at various times of year, as the relative position of the sun changes. “As the sun passes over the scene, different pixels will light up at different times,” says Austin Abrams at Washington University in St Louis, Missouri. The software he and his colleagues created uses a GPS reading, which can be taken separately, and time-stamp data to calculate the position of the sun in relation to a webcam image. By watching how reflections change over the course of a few months, it can figure out the orientation of all the surfaces in the scene.
Google generates the 3D models that populate Google Earth by using a fleet of camera-equipped planes that fly over cities snapping photos. The program also lets users create their own 3D models of local buildings and upload them. Those models are kept simple to ensure that Google Earth runs smoothly. The researchers’ models, by contrast, capture minute detail. “In some cases, we can even capture the 3D structure of individual shingles on a rooftop,” says Abrams.
The new method cannot handle sudden changes in depth, however, such as a building with a mountain behind it in the far distance, because it calculates depth by comparing neighbouring pixels on the webcam images. The best models would result from combining the system’s data with other techniques that are able to more accurately capture broader structures, says Gabriel Brostow, whose team does related research at University College London.
Abrams says the system could also help to automate the study of plants’ life cycles. It is detailed enough to, say, pick up how the wooden structure of a tree changes over time. “We look forward to deploying this algorithm across the tens of thousands of webcams across the world,” says Abrams. He will show the work at the European Conference on Computer Vision in Florence, Italy, next week.