Security gets image-conscious

Video-analysis software is getting better at spotting suspicious activity, but a developing market still warrants caution

Like other large seaports, Canada’s Halifax Port Authority has to juggle competing priorities to protect itself against possible terrorist attacks. Its biggest challenge is monitoring the nearly 14 million metric tons of cargo that pass through the facility each year. Security officials must also protect commercial and private boats and numerous container and fuel terminals on the 241 acres of port authority land that surround the water.

It’s financially impossible to hire enough security employees to patrol the entire deepwater port, but managers say they have an effective alternative. They plan to install a sprawling electronic early-warning system that will use video cameras, radar, motion sensors and shipboard Automatic Identification System transponders to monitor the area.

To ease the burden of analyzing all the data those sensors will generate, the system will use video-analysis software that can alert guards if unidentified ships or individuals enter the area.

“The software significantly enhances our ability to detect and therefore deter someone from attempting some form of a nefarious incident,” said Gord Helm, manager of port security and marine operations at the port authority.

The port authority is completing the first phase of a $20 million electronic-surveillance project, which includes a $10 million command-and-control center with a wall-sized screen to display images enhanced by identification and monitoring software from PureTech Systems.

“The software is important for us because we can drill down into the video picture in the event of an alarm and then be able to initiate a response that is appropriate [for] what we are looking at,” Helm said.

Ongoing challenges
Such complex video surveillance installations are spurring the growth of a worldwide video-analysis market. Topping $60 million in 2005, the sector could reach $400 million by 2012, said Dilip Sarangan, a research analyst at Frost and Sullivan who specializes in security issues.  

However, the market faces challenges in reaching that size. For one thing, the accuracy of analytical algorithms must improve. “This is still very, very much an emerging technology,” Sarangan said. “Only in the last year or two have we started to see companies actually implementing these solutions so they could be used as a complete security solution.”

Mark Denari began evaluating video analytics as a security official at San Francisco International Airport in 2002. He now manages a video surveillance test project as director of aviation security and public safety at the San Diego County Regional Airport Authority. When Denari first considered the software, programs couldn’t distinguish between two people moving through the surveillance area and one person with a shipping box in tow, he said.

“Now the software has started to determine that humans have a nose and appendages,” he said. “Some programs can differentiate hand movements, even if they’re subtle.”

Denari said his test of Vidient software accurately detects trailers, people who wait for an authorized person to open a secure doorway and rush through before the door closes. The software’s analytics are accurate enough to detect if the second person didn’t perform the hand movement of swiping an identity card through a reader, Denari said.

Price is another challenge for the video analysis market. The software costs $2,000 to $6,000 per camera channel for basic motion detection and object identification. Specialized aerial surveillance applications for the military rise to $75,000.

As with many security-related expenditures, agencies often can’t calculate a clear financial return on their investment. “Our return on investment comes in making sure that our passengers are safe, that the screening process goes as smoothly as possible and that our airlines do not incur delays,” said Mason Short, executive director of South Dakota’s Rapid City Regional Airport.

The airport spent about $30,000 for software that sounds alarms if it detects someone trying to bypass security via the exit lane for arriving passengers. “Many airports use a person to watch this lane, but we felt it was the best use of our resources to simply automate the monitoring job,” Short said.

An industry shakeout could be another challenge for the video-analysis market. “The whole video space is extremely crowded,” said Mark Moscinski, vice president at System Development Integration, a systems integrator that specializes in security applications. “I expect there are more vendors now than there will be two years from now. Identifying which companies are stable, have a reliable platform and have resources behind them to be around for the next few years will be a challenge.”

Fewer false alarms
Despite the growing pains, video-analysis software continues to mature. Applications that support proactive threat detection help security officials react to incidents before they happen by spotting suspicious activity. Such software can detect the motion of a person hopping a fence or rushing through a restricted area.

Older motion-detection software was plagued by numerous false alarms triggered by intruders such as large birds or so-called terrorists stalking nuclear power plants that turned out to be wayward deer. Increasingly sophisticated algorithms programmed to mathematically match human height, movement and other characteristics have reduced false alarms.

“We have a database of over 200,000 human figures that our software can match against,” said Steve Goldberg, Vidient’s president and chief executive officer. “It looks for matches between what it sees and any combination of those 200,000 training samples to detect a human.”

“The software constantly compares the changes or lack of changes in the frame,” said Mariann McDonagh, vice president of global marketing at Verint Systems. “The software sends an alert saying, ‘Here is something that is outside of the specifications of what we have deemed as normal.’ ”

Other algorithms detect the distance between an individual and an object he or she has left behind. If a tourist drops a suitcase and then moves 25 feet away, for example, the system alerts security officials to the potentially suspicious behavior.
 
“We’re trying to leverage technology to overcome human failures,” Denari said. “It’s so easy not to notice someone suspicious.”

Video software can also assist with forensic analysis after an incident occurs. Of the three incidents that have triggered alarms since the Rapid City Regional Airport installed motion-detection software in 2006, one involved a passenger who ignored alarms and continued to the boarding area. The video record of the event was a major piece of evidence for the U.S. attorney who prosecuted the breach.

“Without it, it would have been difficult to address the situation,” Short said.

Adding video-analysis software to systems integrations can provide a greater degree of automation for security-conscious organizations. For example, when perimeter intrusion-detection software notices a problem, security systems can automatically lock doors or raise vehicle gates.

However, when such software is used to perform more complex tasks, it often falls short. Facial recognition software — a marquee application for video analysis — still hasn’t seen the success of more basic applications. But now even in this area the software is becoming more accurate, because of a technique called image indexing.

3VR Security uses indexing to categorize facial images in biometric templates that are stored in a database with the original video. The templates highlight specific facial characteristics, such as the shape of a nose or the distance separating the eyes. As security officials try to match two facial images, the analysis software focuses on the highlighted characteristics to narrow its searches.

“It looks at patterns within the face,” said Tim Ross, executive vice president of sales and marketing at 3VR.

The templates are useful for comparing a new image of someone lurking in a parking lot late at night with indexed images to determine if cameras spotted the same person at an earlier time. Rather than have someone laboriously searching through an entire archive, the software can look for a match in the indexed images and produce a results page displaying likely matches in seconds or minutes, 3VR officials said. The software costs about $1,000 per camera.

The company’s facial matching software works only for images captured with 3VR systems or for video files recorded by another camera system and imported into the 3VR system. Ross said indexing a face from a photograph in a watch-list database is difficult because compression technologies make the images unsuitable for video analysis.

A complete picture
When the Halifax Port Authority’s wall screen becomes operational, video input from various sources will offer a complete picture of the port and the surrounding shoreline. The PureTech system will compute the location of each sensor, including its longitude, latitude and elevation.

“Everything will be pulled together into a geospatial system that can help you understand where [a suspicious object] is located and its proximity to the critical infrastructure,” said Larry Bowe, PureTech’s president. Because the electronic security zone will extend beyond the perimeter of the port authority’s fence, cameras will be able to detect movements before someone or something enters the secure area, giving officials time to decide whether to respond.

Many people believe, as Helm does, that terrorists will attack the United States again and the only unanswered questions are where and when. Automated surveillance systems give the public a degree of safety assurance they wouldn’t otherwise have, he said, and they enable public authorities to, at the very least, mitigate the impact of a terrorist incident.

Joch is a business and technology writer based in New England. He can be reached at ajoch@worldpath.com.
Military agencies use video locator assistantsVideo analytics have uses beyond interpreting the output of ground-based cameras. Military and intelligence agencies send video captured by Predator and Global Hawk unmanned aerial vehicles (UAV) to servers equipped with software that detects terrestrial movements or changes in infrastructure.

A version of video-analysis software from Sarnoff lets users search and analyze reconnaissance video captured by low-flying aircraft. “You can fly a UAV over the area of interest — it could be many miles — and identify what changes may have happened from one day to the next,” said Ron Krakower, director of business development at Sarnoff.

A similar product performs real-time analysis and fixes images onto geospatial maps like those provided by Google Earth. “Someone can have a God’s-eye view” as the UAV flies through the air, Krakower said.

One limitation of older aerial surveillance video has been a soda straw view of the world that requires an analyst to look back and forth between the video image and a map to determine the UAV’s position, he said.

“Once you slap the video onto the map in real time, it tells the whole story rapidly,” Krakower said. “And by seeing it in context, you know where that moving object is in relation to the good guys and other assets.”

Real-time analysis software for aerial applications costs $20,000 to $75,000 per video.

— Alan Joch
4 tips for selecting video-analysis softwareBefore selecting a video-analysis application and vendor, agencies should follow these recommendations for creating successful electronic surveillance systems.

1. Evaluate interoperability. “A lot of vendors tout an open architecture and their software’s ability to plug into a variety of systems,” said Mark Moscinski, vice president of System Development Integration. “But when you actually look under the hood, it’s not quite as open as they indicated.” He advises people to choose packaged solutions that bundle as many of the needed hardware and software components as possible. “We’ve had more success where we can try to standardize the platform with fewer vendors,” he said.

2. Look for long-term relationships. Moscinski said a crowded market
doesn’t stay that way for long. In addition to evaluating the software, he said, agencies should consider a potential vendor’s financial history and whether its future rests on a single product or a larger product line that could have longer-term appeal.

3. Conduct test projects. Lighting conditions, traffic patterns, camera models and other factors can affect analytical accuracy. So rather than relying on vendor demos and references, agencies should test software in their environments before making buying decisions. “In some cases, you may need to replace your existing cameras with higher-quality models,” said Mark Denari, director of aviation security and public safety at the San Diego County Regional Airport Authority.

4. Have realistic expectations. Because video analytics is an emerging sector, agencies shouldn’t expect perfect detection rates, said Steve Goldberg, president and chief executive officer at Vidient. However, most applications allow users to adjust their software to better balance detection sensitivity and false alarms, he added.

— Alan Joch

Featured

Reader comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above