Video AI tool watches border crossings
- By William Jackson
- Aug 21, 2003
At the Canadian border, the Homeland Security Department's Customs and Border Protection Bureau is using intelligent video surveillance technology to analyze arrivals.
'We needed a way of identifying intrusions on the border after normal hours of operation,' bureau spokesman Bill Anthony said.
The Video Early Warning system from DiamondBack Vision Inc. of Reston, Va., also known as ObjectVideo, adds artificial intelligence to existing video recorders that monitor many border crossing sites around the clock.
The recorders have tunable motion detection software. 'Anytime motion is detected, you get an alert,' Anthony said. But if the sensitivity is set too high, there are false alarms. If it is set too low, there is a chance of missing something.
VEW analyzes what is happening on the video feed in real time. 'We establish a set of rules, and it notifies us when the rules are violated,' Anthony said.
VEW, which went live in December, grew out of a Defense Advanced Research Projects Agency initiative on computer vision. The software learns the normal conditions of a view being captured by a camera and analyzes any other activity, ObjectVideo CIO Alan Lipton said.Wizard interface
'It can tell the difference between a squirrel running up a fence and a person climbing over,' Lipton said. 'This gives a low false-alarm rate.'
The bureau defines events of interest by entering and updating a set of policies through a wizard interface. Each camera feed can have a separate policy, and there is no limit on the number of rules.
'It's up to the customer to come up with the rules,' Lipton said. 'VEW is not going to give you any alarm until you tell it what you're interested in.'
The software comes preinstalled on a rackmount server running Microsoft Windows 2000 or XP, with Fast Ethernet and Gigabit Ethernet connections. One to four cameras connect directly to its video processor, and up to 20 processors are networked with the server for a maximum of 80 camera feeds.
'It is pretty much agnostic as to camera type, medium and environment,' Lipton said. 'You can distribute the architecture any way you want,' with a central server analyzing multiple remote video feeds.
The server stores up to 4.8T of images in a searchable archive that can serve as a forensics tool for events or objects.
The Border Protection implementation is part of the larger Northern Border Security Development Project, established after the Sept. 11, 2001, terrorist attacks.
VEW performed a one-week pilot test in January at four locations in Washington state. For security reasons, the bureau is tight-lipped about how much it has spent on the project and how many VEW systems are being installed where.
'Several,' is all Anthony would say. But he added, 'In the future, all land ports of entry will have them.'