"Questionable Observer Detector"
identifies people who keep popping up in crime scene footage

Chances are, you've seen at least one or two TV shows in which the police examine news footage shot at several different crime scenes, and recognize the same person's face showing up repeatedly in the crowds of onlookers ... the ol' "criminal returning to the scenes of their crimes" scenario. Realistically, it's pretty hard to believe that one person could look through all that footage, and remember all those faces. It turns out that a computer could do it, however, as scientists at Indiana's University of Notre Dame have illustrated with their "Questionable Observer Detector," or QuOD.

The system was developed by a team led by Kevin Bowyer, Patrick Flynn and Jeremiah Barr, of Notre Dame's Department of Computer Science and Engineering. Bowyer was inspired to invent QuOD, when he heard military and national security experts describing how they needed a way of identifying IED bombers in the Middle East - such individuals are known to come back to bombing sites, to assess the aftermath of their work.

While most facial recognition systems match faces in footage to those in an existing database, there is no such database for QuOD. Instead, it creates a separate "face track" for each individual appearing in a video clip. For each new clip that it analyzes, it compares the face tracks created from that clip with those created from all the previous clips. Any time that matching tracks are discovered, they are grouped together so that a human operator can then view all the appearances of that one person.

One of those images could then possibly be matched to a photo in a database, or at the very least circulated in the form of a "WANTED" poster.

Although it has already been successfully tested on a relatively small scale, there are still a few challenges to be overcome before QuOD can become a practical tool. For one thing, the lighting and resolution of crime scene footage is often insufficient, and not all of the onlookers will be conveniently looking towards the camera. Also, the processing load on computers could become quite heavy, if large amounts of footage were being analyzed.

Nonetheless, the Notre Dame team believe that these problems can be solved, and that QuOD could one day be used to catch criminals who just have to come back for a second look.

The video below shows the system in use.