New algorithms can tell where videos were filmed

Researchers have developed new algorithms which can locate where a video was filmed from its images and sounds.
In the future the algorithm could help to find people who have gone missing after posting images on social networks, or even to recognise locations of hostage executions and operations of terrorist groups, researchers said.
Many of the videos available online are accompanied by text which provides information on the place where it was filmed, but there are others that do not present this information.
Scientists from the La Salle campus at Ramon Llull University (Barcelona) have developed a system to locate videos with no indication of where they were produced on the map, a real challenge considering that the majority are scenes of daily life without the appearance of clearly recognisable places.

As they do not come with text, the method is based on the recognition of their images or frames and all of the audio.
"The acoustic information can be as valid as the visual and, on occasions, even more so when it comes to geolocating a video," said Xavier Sevillano, one of the study authors.

"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," Sevillano said.
All of the data obtained is merged together and grouped in clusters so that, using computer algorithms developed by the researchers, they can be compared with those of a large collection of recorded videos already geolocated around the world.

In the study, published in the journal Information Sciences, the team has used almost 10,000 sequences as a reference from the MediaEval Placing task audiovisual database, a benchmarking initiative or assessment of algorithms for multimedia content.
"The videos which are most similar in audiovisual terms to what we want to find are searched for in the database, to detect the most probable geographical coordinates," said Sevillano.
Sevillano pointed out that the proposed system "despite having a limited database in terms of size and geographical coverage, is capable of geolocating videos with more accuracy than its competitors."
More specifically, it is capable of locating 3 per cent of videos within a ten-kilometre radius of their actual geographical location, and in 1 per cent of cases it is accurate to one kilometre.

The percentages are still modest, although they are four times more precise than the accuracy reached up until now using the same database, researchers said.