The questionable observer detector

January 25, 2011 by Lakshmi Sandhana

(Image: University of Notre Dame)

Exclusive | Kevin W. Bowyer, Chair of the Department of Computer Science and Engineering at the University of Notre Dame, Indiana is out to create a tool to reliably identify criminals who may be hanging out at the crime scene after the event.

Their Questionable Observer Detector (QuOD) can process any available video clips of groups of people present at the scene of event, spanning different times and locations to pick out any person who appears frequently in them. “The idea is that the person showing up unusually often as part of the crowd at these events may be someone that the police would want to talk to,” says Bowyer.

For instance, if given video clips of the serial bomb attacks currently plaguing Abuja in Nigeria, the software could identify people appearing repeatedly, giving authorities leads on either the bombers or their helpers.

It’s an extremely challenging 3D facial recognition problem, since the system has no prior database to compare the faces with and works with videos taken by surveillance cameras and bystanders with their handheld video camera under any and all conditions. It detects potential suspects by creating individual “face tracks” for people appearing in every video, comparing them across videos to zero in on similar looking faces.

It takes the best image from a face track in one video and compares it with the top images derived from other face tracks in other videos to see if it can find the same face. When it spots a match, it adds it to a group of video appearances featuring just that person. It identifies an individual as a “questionable observer” if the final number of videos in their personal group exceeds a particular number, set earlier by authorities. This indicates that they appear often at numerous crime scenes.

(Image: University of Notre Dame)

A research prototype could be ready in less than a year, according to Bowyer, who hopes to work with the Federal Bureau of Investigation (FBI) and the Intelligence Advanced Research Projects Activity (IARPA) later on.

Is the QuOD a step further towards the dark picture painted by George Orwell in his novel 1984 of omnipresent surveillance and pre-emptive arrests? Given that it is more about handing authorities a tool to identify suspects based on their actual presence solely at multiple disaster areas, as opposed to policing people on possible intent (as in the movie Minority Report), the answer would have to be “no.”

(Image: University of Notre Dame)