Math professor helps pioneer automated way to spot threats in large crowds

Engineers at the University of Central Florida are devising a way to use video to detect abnormal behavior in large crowds in an effort to help keep people safe from terrorist attacks.

Professor Mubarak Shah heads the team that is using video streams, algorithms and a unique analysis to figure out how crowds behave and what might be suspicious behavior.

Shah’s team includes Brian Moore, an assistant professor of mathematics at UCF; Saad Ali, a UCF Computer Vision Lab alumnus and a computer scientist at SRI International in Princeton, N.J., and Ramin Mehran, who completed his Ph.D. last month at UCF and is joining Microsoft in January.

Video cameras monitoring people in public areas are now commonplace, and in largesettings such as the Super Bowl, political rallies or royal weddings, the stakes are even higher because there are more people, which means more potential casualties.

Several research efforts are under way to develop systems that cue security personnel to individuals or events of interest in crowded scenes. Most of that now ishandled by people who can easily miss something because it is impossible for them to monitor every change in the midst of a large crowd.

Shah is trying to automate most of the work with computer algorithms, and he’s modeling his work on how liquids behave in motion.

“While human psychology and individual quirks complicate the analysis, in essence people in high-density crowds appear to move with the flow of the crowd, like particles in a liquid flow,” Shah said.

So his team is building a program that analyzes behavior in a crowd much like analyzing particles and how they act in a fluid state. But there is one variation.

“We are saying one difference between crowds and fluid is that people have somedestination they are going towards. That’s why crowds can be called ‘thinking fluid,’” Shah said.

The team’s work is promising and was the cover story last month in Communications of ACM, a computer science periodical.