SAN FRANCISCO, Calif.--Surveillance firm 3VR announced on April 15 that its facial recognition video platform demonstrated the highest percentage of accuracy in a test of biometric and video security solutions conduced by the South Korean National Police Agency.

During a two-year testing phase, 3VR's SmartCams, megapixel cameras with integrated 3VR intelligence, performed between an 85 percent and 92 percent accuracy range in recognizing and matching faces. The tests were conducted in a high-traffic public train station, where it was "not uncommon for 100 people to get off a training in a one-minute or two-minute period," said Steve Russell, chairman and founder of the company.

The South Korean government embarked on a project to test biometrics after the country experienced a number of high-profile kidnappings. In one instance, the President of the country tasked thousands of police officers to review millions of hours of government surveillance video in an effort to find a missing child. Unfortunately, the initiative failed to produce any concrete leads and the child was found dead, propelling the government to look for a solution that would enable them to search through vast amounts of video data for investigative purposes. SKNetworks, one of the largest telecommunications and security providers in the country, led the initiative.

Russell said 3VR's platform is able to demonstrate a high level of accuracy because it was designed specifically to work with video.

"Every other facial recognition product on the planet started with the idea of comparing something akin to a flat, normalized, pose-controlled passport photo," he said. "Any algorithm that emerged or has a deployment history in that market is fundamentally ill-suited for use in surveillance video. Video is a different beast."

Why? The number of images that need to be processed is higher and faces are smaller, for example.

The second issue that separates 3VR's solution from others, Russell said, is that it was designed to be able to handle video from thousands of cameras. "That ability to scale becomes extremely important when you have circumstances where tens of thousands of people are streaming by a camera," he said.

Prior to South Korea's research study, a similar test was conducted by the German Federal Police in 2006. This survey of facial recognition technology was conducted under controlled circumstances to limit environmental effects and the best accuracy rate recorded was 60 percent.

Now that the South Korean study is completed, the NPA's goal is to deploy a much larger facial recognition system across Seoul and beyond, Russell said.

This is by no means a done deal for 3VR -- the government is now in the process of choosing a general contractor for the project, which will also include cameras, command centers and emergency call facilities. When completed, it is expected to be the largest deployment of facial recognition in the world.

"Our expectation is that the general contractor will bid the best technology and this study will be used to inform," Russell said. "When it comes down to the core platform that is used to actually perform the biometric extraction and analysis, our expectation is that no other technology can perform as powerfully as we did. Ours was the only technology that met the NPA's needs in terms of accuracy and scale."