'Eyesight' software will scan MARTA for suspicious behavior

Can a computer program catch a bad guy before he even commits a crime?

MARTA will soon see.

The transit agency has contracted with a Houston-based security firm to analyze its surveillance video using an artificial-intelligence-based software that “learns” to distinguish normal behavior from that which seems suspicious. Much like a human watching a camera for weeks would do, it begins to understand patterns over time and picks out anything that seems anomalous.

The program then sends an email alert to MARTA police communications dispatchers about potential dangers. The alert includes a video clip of the suspicious activity. Such activity could include a passenger walking onto the tracks, leaving a bag, breaching the fence line or entering an off-limits area.

“We can’t be everywhere, but this is another tool in our tool kit to help customers feel safe,” said MARTA Police Chief Wanda Dunham.

Behavioral Recognition Systems (BRS Labs) has patented the technology for its AISight (pronounced “eyesight”) software and has other projects in development in Washington, D.C., Chicago, San Francisco, Los Angeles and New York. It is also working with Amtrak, according to its president, John Frazzini.

MARTA bought the software, hardware and three years of maintenance and support for about $4 million. Of that amount, $2 million was funded by a federal transit security grant program. MARTA’s board approved the purchase in May 2013.

The technology will be phased in with stationary cameras in all 38 MARTA stations over the next year and will serve as a force multiplier for MARTA police at each concourse, Dunham said. MARTA will be evaluating the system on an ongoing basis, a spokeswoman said, but no specific measure or time frame was mentioned for determining if it is a success. Eventually, MARTA hopes to expand the software to analyze the footage from cameras in its trains, buses and parking lots.

“We have over 2,000 cameras on our system, and we do not have someone watching those cameras 24 hours a day,” said Sgt. Aston Greene, commander of MARTA’s emergency preparedness unit. “In the past, we only had the ability to react. This solution gives us the ability to prevent something from happening.”

The idea of a computer watchdog seems potentially invasive to some people, among them Paul Cooper, a technology blogger for ITProPortal.com.

“In my view, handing over detection of crimes to autonomous systems has Orwellian overtones that should make us all pause,” Cooper said in an email Thursday.

But Frazzini said such concerns are needless because the program is based strictly on analyzing behaviors, not people. The program does not collect any identifiable information about an individual’s likeness, size or race. It doesn’t use facial recognition software or license plate readers, Frazzini said.

“The use of technology often has a tendency to create concern related to privacy,” Frazzini acknowledged. “But we think our approach strikes a great balance between privacy of individuals but providing security as well.”