By Mark Ingebretsen, HSO Contributor
When Super Bowl crowds converge on Miami's Land Shark stadium this February 7, they'll be met by an army of law enforcement personnel that's been in place for more than a month. As the Associated Press reported last December, an entire section of the stadium's parking lot contained "armored SWAT vehicles, bomb-handling robots, mobile command posts, explosive-sniffing dogs, even a large X-ray device used by the US Homeland Security Department to screen ocean-going shipping containers." The manpower and gear was intended to guard the Orange Bowl and Pro Bowl games, held at Land Shark in the weeks prior to the Super Bowl. And it was designed to send a not-so-subtle message, the AP says, "anyone, terrorist or otherwise, plotting an attack on any of these high-profile events had better think again."
By This e-mail address is being protected from spambots. You need JavaScript enabled to view it HSO Contributor
When Super Bowl crowds converge on Miami's Land Shark stadium this February 7, they'll be met by an army of law enforcement personnel that's been in place for more than a month. As the Associated Press reported last December, an entire section of the stadium's parking lot contained "armored SWAT vehicles, bomb-handling robots, mobile command posts, explosive-sniffing dogs, even a large X-ray device used by the US Homeland Security Department to screen ocean-going shipping containers." The manpower and gear was intended to guard the Orange Bowl and Pro Bowl games, held at Land Shark in the weeks prior to the Super Bowl. And it was designed to send a not-so-subtle message, the AP says, "anyone, terrorist or otherwise, plotting an attack on any of these high-profile events had better think again."
 |
While security officials are understandably coy about what anti-terrorism tools they'll use at February's big game, one of the technologies which may be in place consists of software algorithms designed to automatically monitor the audio and video feeds from multiple surveillance cameras, and then flag suspicious activities in real time. Such algorithms have been under development for several years in the US and Europe, and besides their potential benefit in thwarting terrorists, they're also being used by law enforcement to safeguard neighborhoods, entertainment venues and prisons.
One of the most ambitious efforts is an EU initiative called ADABTS (Automatic Detection of Abnormal Behavior and Threats in Crowded Spaces), designed for use in large public areas such as stadiums and airports. As the program's prospectus describes it, "ADABTS aims to address one of the key problems, the definition of abnormal behavior." Specifically, the video analysis algorithms must understand what constitutes suspicious behavior before it can sound an alarm. To accomplish that, human analysts must recreate crowd behavior, then teach AI algorithms to recognize instances that should be flagged.
Screams in the elevator
Getting computers to recognize visual cues is a daunting task. Even state-of-the-art robots must be painstakingly trained to recognize common objects such as tea cups and telephones--which is why some researchers have opted instead to develop video surveillance systems that are linked to microphones.
At Portsmouth University in the UK, for instance, engineers are working on a system that uses fuzzy logic to first identify suspicious sounds, then automatically rotate a video camera toward the source.
Sound identification programs usually work by matching real time audio signals with a library of noises designed to mimic suspicious activities. Statistical analysis programs then determine the probability of a match. For example, when scientists at Mitsubishi Electric Research Laboratories in Cambridge, Massachusetts, began work on an office-building monitoring system, they first recorded noises such as the whir of copying machines or people chatting that you might expect to hear in the workplace. These sounds served as baselines, explains Bhiksha Raj, a researcher at the facility who spearheaded the project. Next, the group hired actors to play the part of someone being mugged in an elevator, in order to create a library of anomalies.
Even correctly identifying sounds is far more difficult than it appears, notes Tjeerd Andringa, a physicist by training, who is co-founder of the Dutch start-up, Sound Intelligence. Just one of the challenges, for instance, is that sounds don't emanate from a single source. Instead, they ricochet off a multitude of surfaces, which can drastically alter the sound waves' characteristics. Sound Intelligence's solution is to simulate how humans perceive sounds by creating a linear model of the inner ear. The algorithmic model is thus able to identify aggressive noises such as shouts, without having to rely on a large library of examples.
Mitsubishi researchers use another way to identify suspicious sounds, by analyzing how they appear in sequences. A gunshot-like noise could result from any number of sources, a car backfiring, for example. But gunshots preceded by shouts would be a more likely sign of trouble.
Used in venues such as Land Shark stadium, systems like those being developed at Mitsubishi will be called upon to interpret feeds from hundreds of sources. And in the short term, those systems will be hard pressed to duplicate what humans can readily do. By some estimates, a person placed in a crowd can react to a suspicious word, gesture or noise in just 300 milliseconds.
Mark Ingebretsen is a free-lance journalist specializing in science and technology.
Subscribe to this comment's feed
| < Prev | Next > |
|---|
Company