You know what they say about returning to the scene of a crime, right? Some criminals just can't stay away, and a team of researchers at Notre Dame University is using computer vision to capitalize on that urge.
Their Questionable Observer Detector system is designed to identity suspicious people who gather at crime scenes. The software will pick up on those who repeatedly appear in bystanders' videos, even if the quality is amateur-grade.
"QuOD" creates "face tracks" for each individual in a video, repeating the process for all available video clips. When the face tracks are compared, a cluster is created of any faces from different video clips that look similar enough to match each other.
An individual is considered suspicious if law enforcement considers that he or she appears too often for comfort.
There are still problems to overcome - lighting and resolution for such bystander videos are often very poor indeed, and the sheer number of videos to be processed can often cause difficulty. But the team say they're confident of overcoming these challenges.
Far more important for many, though, will be the issue of civil liberties. The researchers say they believe people will approve, given that the system identifies people who were actually present at crime scenes, rather than simply suspected of being there.