The completely automated system creates thousands of time-lapse videos using crowdsourced images from the internet
Researchers from the University of Washington have developed an approach for creating time-lapsed photos--often spanning a decade--using nothing more than publicly-sourced images taken by Internet users.
The Internet, being a natural repository of billions of digital images taken by users over decades proved to be a treasure trove for the researchers in putting their AI-based image recognition system to the test. By smartly applying image manipulation processes that counter variations in viewpoint, lighting conditions and image quality, the system is able to automatically identify and stitch together relevant photos of a given location into a seamless whole.
These videos portray phenomena that otherwise take months and even years to pass; movements of glaciers, changes in river courses, constructions of skyscrapers and the like. Sped up by a factor of about 10 million, these glacial movements become living sequences that lend an altogether different perspective as these phenomena unfold.
The approach, completely automated, consists of image recognition from a cluster of 86 million photos from popular public image sources like Picasa and Panoramio, grouping them into popular landmarks and similar viewpoints. These subsets are then sorted by date, and tweaked to conform to a common viewpoint. The effect is finished off with filters that compensate for variations in lighting and reducing flicker while creating the final time-lapsed blend.
"A typical time-lapse with 1,000 input posed photos takes about six hours to compute on a single machine," say the engineers. "While the algorithms can be optimized a lot more for efficiency, we point out that a few hours is negligible compared to the time period of several years it took to capture the photos.”
Watch here:
If you happen to be interested in deep diving into the science behind this development, find the research paper here.
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
)
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
)
){kind=small}
){kind=small}
)
)
)
)
)
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}