ACES profile: The face in the crowd

Any fan of film thrillers recognises the scene. Working against the clock and the odds, police officers scan the thousands of faces pouring through the airport or the train station or the busy downtown shopping area, trying to identify and apprehend their suspect before it’s too late. After building the tension to breaking point, the film will have the police arresting their man, usually thanks to the singular wits of the hero.

The researchers behind Kee Square Srl, the winner of the 2008 Academic Enterprise Award for IT, think the police should use a computer. The company was spun out of the Politecnico di Milano with commercial ambitions for software that can help security services spot the face in the crowd.

The company had security applications in mind from the start, said Stefano Tubaro, one of the founders of Kee Square. “The principal sector is the public sector: airports, train stations, police bodies.”

Tubaro, a professor of signal processing at the Politecnico, working with his colleagues, Augusto Sarti and Pasquale Pigazzini, joined forces with the Politecnico, Celin Technology Innovation S.r.l. and Ghirlanda SpA in July 2007 to form Kee Square. Quantica SGR, a Milan venture capital fund, came on board in December 2007, bringing €1.5 million in capital along with it.  

Facial recognition and classification technology isn’t new. But Tubaro says that what he and his colleagues did was to make the software far easier to use without buying expensive hardware on which to run it.

“Other solutions require more hardware to run,” he said. “One of the problems of the systems is the cost. Even in the public sector, there is a need to reduce costs.”

In only its second full year of existence, 2009, Kee Square should generate revenue of €1 million, double the 2008 amount, Tubaro said, adding that the company also expects a slight profit in the current year. In three years, the company hopes to deliver revenue of €10 million, although Tubaro leaves plenty of room for variation, depending on economic developments and the success of Kee Square products outside the security sector. The company spends about half its revenue on research and development, he said.

Think of a group of people waiting for a plane, Tubaro said. They’re standing in front of an advertising screen. The advertiser doesn’t know which ones are watching the screen, for how long, and with what kind of response. “There are a lot of these screens in these public spaces. There is no information on how efficient that type of advertising is.”

Kee Square thinks its technology can classify the responses and help the advertiser know what’s effective and what isn’t. The leader of a focus group meticulously tries to gather such responses by asking questions and noting the responses. With the right software, Kee Square believes the camera can study the audience in its natural setting and deliver fast feedback to the company.

We are trying to enlarge our software products to understand human activities in a more general way, Tubaro said. Kee Square hopes to help develop smart cameras that would digest the information even as they gather it. Security information may have to be sent on to a central processing station, but smart cameras could do much of the analysis for advertisers, at least in some cases, he said.

Tubaro estimated the advertising use of the software could account for as much as 20 per cent to 30 per cent of company revenue in 2009.

Unsurprisingly for researchers coming from the image and sound group at the Politecnico, Tubaro and his colleagues are developing sound systems analogous to its work on images. Security forces are again the primary target market.

“There are events that aren’t so easy to detect using images,” Tubaro said.

Security cameras won’t see when a gun goes off, or glass breaks, and certainly not voices being raised. Tubaro said Kee Square’s work on sonic technology uses microphones placed in public places to help police identify what can’t be seen on camera. In the best scenarios, the sound may alert the police to trouble before it happens, he said.

“We think the use of images and sounds can increase in a significant way the performance of surveillance systems,” he said. Raised voices, for example, could alert security when a “normal conversation degenerates to something that is not so normal.” Video security comes after the fact. But sound can allow detection in real-time or even slightly in advance of a problem, he said. Kee Square’s sound system is currently being tried in a Naples underground station.

From software that’s meant to help security services pick a face out of a crowd, it’s not a huge step to other face-recognition applications. Kee Square’s products can take pictures that meet the standards for electronic identity documents or control physical access to a site.

From a successful career in research and publishing, Tubaro found the adjustment to practical application the biggest surprise in his new commercial activity. Thinking about problems related to algorithms and then adjusting one’s thinking to the challenges of a real company, that’s a big step, he said. “It is different from developing an idea or publishing in a scientific journal.