New Algorithms Locate Where a Video Was Filmed From Its Images and Sounds

Creation of A System to Geolocate Video

A system capable of geo-locating video on comparing with their audio visual content with worldwide multimedia database has been created by researchers from Ramon Llull University in Spain for instances when textual metadata is not relevant or available. This could be helpful in the future, in locating people who could have gone missing after posting images on social networks or to recognize areas of terrorist activities. Several of the available online videos have text providing information on the destination it was filmed though there are others which do not provide the information which complicates the application of the frequent geo-location tools of multimedia content.

To resolve this issue, scientists from the La Salle campus at Ramon Llull University – Barcelona - Spain, have come up with a system which helps in locating the videos without any indication of the location where they were produced on the map, which is very challenging, taking into account the major part of the scenes of daily life minus the appearance of clearly recognizable places. Since they are not accompanied with text, the system is based on the recognition of the images or frames and the audio.

Physics & Mathematical Vectors Taken

One of the authors, Xavier Sevillano, states that `the acoustic information can be as valid as the visual and on occasion even more when it comes to geo-locating a video. In this field the use of some physics and mathematical vectors is taken from the field of recognition of acoustic sources because they have already demonstrated positive results’. Data gathered is put together and diverted into clusters, to enable using computer algorithms that has been developed by researchers and compared with a large collection of recorded video that have already been geo-located across the world. The team in their study published in the journal - `Information Sciences’, used around 10,000 sequences as reference from the MediaEval Placing task audio-visual database which is a benchmarking initiative or assessment of algorithms for the purpose of multimedia content. Sevillano states that `the videos that are similar in audio-visual terms to what they need to find, are searched for in the database, to identify the most probable geographical coordinates’.

Recognized the Need for Greater Audio-Visual Base

The researchers also indicate that the proposed system inspite of a limited database with regards to size and geographical coverage has the potential of geo-locating video accurately than its competitors. Moreover, it is also capable of locating 3% of videos within a radius of ten kilometres of its actual geographical location and in 1% of the cases; it tends to be accurate to one kilometre.

The researchers have also recognized that their system may need a greater audio-visual base in order to apply it to millions of videos that tend to circulate across the internet though they highlight the usefulness in locating those which do not have textual metadata as well as the capable possibilities it offers. Sevillano further states that `this system could help rescue teams in tracking down where a person or a group may disappear in a remote area and detect the locations portrayed in the video that could have been uploaded in a social network before they lost contact.