Biometrics -- facial recognitionIdentifying faces in a crowd in real-time

Published 2 September 2010

U.K. company develops a face recognition technology that can recognize individual faces in a crowd — and do so in seconds, even when they are moving, at a wide angle, or in poor light; the system captures and analyzes images and compares them to a database, and alerts security personnel if a match is made

A new type of infrared security camera is being used to identify individual faces in real time. The CheckPoint.S system from Guildford, Surrey, U.K.-based OmniPerception can recognize faces in seconds — even when they are moving, at a wide angle, or in poor light — by capturing and analyzing images and comparing them to a database.

As well as alerting security teams to a suspect’s presence, the system can also be used to confirm the identity of invited guests entering a building. “The thing that has plagued facial recognition for many years is the effect of lighting,” OmniPerception’s CEO Stewart Hefferman told the Engineer’s Stephen Harris. “Our systems operate in the near-infrared space rather than visual light. This means we effectively provide our own illumination source and we don’t care if the system is operating indoors, outdoors, in a bit of sunshine, mottled sunshine or whatever — we can cope with that.”

OmniPerception’s cameras emit near-infrared light, which is invisible to the human eye, and then use the reflected waves to scan the subject’s features. If the face matches one on record, the CheckPoint.S system immediately alerts the operator.

The system does not require constant human monitoring and is not intended as a replacement for CCTV cameras, which usually look down on a location rather than filming at eye level to record people’s faces.

Hefferman stressed the technology was designed as a tool to alert security teams or police to the possible presence of an individual, rather than a watertight way of proving someone was at location at a specific time.

We may not recognize someone with 100 percent confidence and that’s why we always pass the final call to a more intelligent being — a human — and we choose the applications and markets on that basis,” said Hefferman. “We would give the police the confidence to pursue a line of enquiry rather than the evidence to say ‘John Smith was there.’ I don’t ever believe it would be used purely as an evidential tool.”

Harris notes that the firm has been using infrared cameras for around six months but more recently introduced the algorithms that allow CheckPoint.S instantly to identify its subjects. It was developed over eighteen months, partly in collaboration with BAE Systems, which helped perfect the technology’s greater tolerance to scanning faces at wide angles or when moving.

The U.K. government indicated in its coalition agreement that it intends to introduce legislation to regulate CCTV (“U.K. will regulate license number plate recognition cameras more tightly,” 6 July 2010 HSNW). Hefferman said he would welcome regulation as a way of addressing public concerns about security systems and invasion of privacy.