Paul Howie, NEC Australia's general manager of Smart Systems, said the main expense would be on cameras and hardware. There would also be costs involved for software and algorithms that do the actual recognition.
The Australian federal and state governments today agreed to set up a national facial recognition database, using photos from drivers' licences and other data.
Howie was confident that his company's algorithms would identify people with more than 99% accuracy. "Trials with NIST in the US achieved a 99.2% accuracy rate," he said, adding, "when it comes to both speed and accuracy, we're number 1."
Howie was upfront about privacy concerns but said decisions on such matters had to be made by society, and not technologists.
A great deal of progress had been made in facial recognition technology since the Boston Marathon bombings in 2013 when there were difficulties in identifying people in the crowd, he claimed.
In Boston, facial recognition was attempted after the fact. Now it was being done well in advance. "Say you have 100 cameras in a stadium, then before an event you watch the crowd from a biometric perspective," he added.
Such systems were already in use in other countries, he pointed out; in the US, for example, the federal government and the states had an agreement to share data. NEC has set up facial recognition systems in 40 countries.
Some countries used aggregated systems, while others opted for disaggregated systems; in the latter case, all the data in the distributed systems were not present in the central database.
"This may be needed, say if there are people who are in witness protection programmes," Howie said.
Asked about issues of image quality and image size, which have proved to be stumbling blocks to accurate identification in the past, he said the technology being used now had improved to the extent where a photo from any smartphone was sufficient to provide enough points for identification.
Regarding the angle from which photos were taken, which can often influence the recognition score, Howie said NEC's algorithm worked in two phases: it picked up details from a number of angles and then reconstructed a 3D image for better resolution and visibility.
Regarding costs, he said a stadium the size of the MCG would require expenditure of between $1 million and $1.5 million. A lot of the expense would be for pulling feeds from existing cameras and directing them into a central database.
A system such as the one likely to be required by the Australian government would only be installed in public places like airports and stadiums, Howie said.
As as far as NEC systems were concerned, the company provided processing and storage of images in government data centres behind their firewalls, he said in response to a query.
NEC is currently working with the Australian Criminal Intelligence Commission — a merger of a number of government agencies — to replace its existing fingerprint database and to set up a facial recognition system.
Its competitors in the field are Morpho and Cognitech. The company has already licensed its facial recognition algorithms to a company known as Vision-Box that is setting up arrival and departure gates in airports around Australia.