Faces In The Crowd

In a test conducted for the Virginia Beach Police Department during the summer of 2002, facial recognition software found and matched faces in a crowd 87 percent of the time to facial images stored in a database.

As a result of the test, Virginia Beach upgraded a 10-year-old CCTV system that monitors the city’s beachfront entertainment district with FaceIt ARGUS, a facial recognition system supplied by Identix Inc., Minnetonka, Minn. Funds for the upgrade came from $200,000 in state and city grants.

In operation since the end of last summer, the system compares the faces it scans with 2,500 facial images stored in a database. The stored images include runaways, missing persons, and people with outstanding felony warrants. So far, the system has not encountered a person whose image is stored in the database.

Virginia Beach has been moving toward facial recognition CCTV surveillance since police officials saw a demonstration of the technology in 1999.

While facial recognition technology has been around for a number of years, its use as a surveillance tool is relatively new.

Since the mid-1990s, the technology has served access control applications by matching the saved image of a PIN user or access cardholder to the face of a person requesting access.

Facial recognition has also proved useful to ensure against duplications within large databases. For example, departments of motor vehicles use the technology to prevent individuals from acquiring more than one driver’s license.

Both of these applications are simple compared to that of surveillance, according to Frances Zelazny, a spokesperson for Identix. Surveillance software must find faces in crowded scenes, extract the images of those faces, and compare them to images held in a database.

The process requires enormous computing power. A particular video scene, for example, may contain images of dozens of people, along with fire hydrants, storefronts, automobiles, dogs, streetlights, and many other things. Surveillance software, such as FaceIt ARGUS, must distinguish faces from other images in the scene and extract the facial images one by one to create a digital map of each. The software maps facial images by measuring the relative distance between points on the bone structure of a face and creating a detailed template. Finally, a search engine compares the template to templates stored in a database. Upon finding a match, the system alarms.

One to two officers have monitored the Virginia Beach CCTV system from eight to 16 hours a day, depending upon the time of year, since 1993. According to Mullen, the addition of surveillance software has required no increase in staffing.

Since last summer, the Virginia Beach system has alarmed regularly. In each case, the system operators evaluated whether the match suggested by the computer warranted investigation by police officers. In one case, officers were dispatched, but it turned out to be a false alarm.

“That proves the system works,” Mullen says. “For most of the possible matches, the system operator looked at them and decided they weren’t close enough to send out an officer.”

The real test will come this summer, according to Mullen. A resort destination, Virginia Beach attracts as many as 3 million visitors during the tourist season. While not all of those people will visit the entertainment district monitored by the system, many will.

Noting a budding privacy controversy concerning the technology, Virginia Beach has gone to great lengths to address these concerns. The police department held a series of public meetings to answer questions about how the system would be used. “We brought in experts on both sides, from the ACLU (American Civil Liberties Union) and the Rand Corporation,” says Mullen. “We also met with civic leagues.”

The public discussion raised four issues:

• How long would surveillance images remain stored in the system?

• Would the system be connected to other, larger law enforcement systems?

• How would the department disclose the use of the system?

• How could citizens be sure the department was following policies and procedures acceptable to the community?

According to Mullen, the department had never stored CCTV video and did not intend to start doing so following the addition of surveillance software. Mullen adds that the department does not intend to connect the new system to any other system. Third, the department has posted signs informing visitors that “smart video cameras” are in use.

Addressing the fourth public concern — community oversight — the city council voted to approve using the system. In addition, the police department formed an audit committee to act as a citizen’s advisory group — helping to develop and oversee policies and procedures related to the system. The group includes representatives from the NAACP, the American Filipino Association, the Hispanic community, the business community, the Human Rights Commission, and the Council of Civic Organizations. A citizen volunteer also sits on the committee.

This summer, while FaceIt ARGUS looks for faces in the crowd across the city’s entertainment district, the audit committee, through regular inspections, will keep an eye on how the police are using the technology.