Software could determine where a video was shot, based on scenery and ambient sound
Sometimes, a posted video is the only clue to the whereabouts of a missing person, or a terrorist group. Unfortunately, unless that video has already been geo-tagged, it can often be very difficult to tell where it was shot. Now, however, scientists have created algorithms that can determine a video's location by comparing its background imagery and audio to that of thousands of other videos.
The researchers, from Spain's Ramón Llull University, utilized the existing MediaEval Placing Task database. It's a collection of videos and photos from known locations, used for developing geolocation software.
Given that the videos being scrutinized presumably wouldn't show well-known landmarks, the algorithms instead look for distinct images in seemingly generic scenes (such as streets or wilderness areas) that have a match in the database.
Random ambient audio also plays a key role. "The acoustic information can be as valid as the visual and, on occasions, even more so when it comes to geolocating a video," said researcher Xavier Sevillano. "In this field we use some physics and mathematical vectors taken from the field of recognition of acoustic sources, because they have already demonstrated positive results."
So far, the system has been tested by comparing sample videos to approximately 10,000 other clips already in the MediaEval database. It was able to geolocate three percent of those samples to within a 10-km (6.2-mile) radius of where they were actually shot, and one percent to within one kilometer (0.6 mile).
While those figures obviously leave a lot of videos still unlocated, the scientists believe that the success rate should be much higher if the database is expanded to include many more clips. Additionally, even in its present form, the technology is claimed to be more accurate than other approaches used on the database.