Dubai Police Use Real-time Image Processing in Traffic to Track Down Scofflaws, Terrorists 

When people drive, their cars identify who they are. Or who they might be. This includes people “of interest” to police: drivers with expired licenses or unpaid fines, with outstanding arrest warrants, who may be sought in a criminal investigation or who are driving cars that have been reported stolen. Or even people whose names appear on terrorist watch lists. In short, any vehicle with a license plate can serve as a tool for law enforcement.

The problem for police, of course, is that there are too many cars on the road to track them in anything approaching real time. Until now. In Dubai, one of the seven United Arab Emirates (UAE), police have implemented an Automatic Number Plate Recognition (ANPR) system installed on 150 police cruisers. The system includes mounted ruggedized cameras that capture the numbers of two license plates per second, compares them to data for wanted vehicles held in a central repository, and then alerts officers when a car of interest has been spotted.

The system has delivered impressive results. In one 18 month period, the system helped Dubai Police apprehend 2,739 people locally and internationally.

ANPR was developed by Zenith Gulf Security Systems and integrates HPE’s IDOL, a search and analytics engine that delivers actionable intelligence from both structured and unstructured data. In determining requirements for the ANPR system, Dubai Police wanted frequent updates to the data replicated to patrol cars, which carry the license plate data for vehicles that police are interested in. The main HPE IDOL repository is maintained at headquarters.

The ANPR system has a pan-tilt camera device mounted above a patrol car’s light bar. It was designed primarily for scanning parked cars in garages and lots, but also can scan license plates immediately in front or behind a moving patrol car.

When a car of interest is identified by the system, “the display on the officer’s dashboard shows a picture of the suspect vehicle as well as the highlighted plate data, plus a voice message as part of the alert,” said Russell Hammad, CEO of Zenith Gulf Security Systems. “A screen comes up, shows you the image of the car, as well as all relevant records pertaining to the vehicle and why it has been identified.”

HPE's David Humphrey

“The trick here is you’ve got significant (compute) resource requirements actually in a vehicle,” David Humphrey, CTO video and imaging, HPE Big Data, told EnterpriseTech. “That’s where you’ve got these emergent technologies, between surveillance video analytics and with the movement toward Internet of Things, where you’re pushing number crunching out to the edge. So you distribute tasks such that you then get the sensible answers back to your central control site for identifying the vehicles and the license plate numbers, but you’re not pushing video across the 4 gigs of bandwidth.”

Frequent updates to the main repository – and to the actionable data in police cruisers – is important so that drivers who have paid their fines or resolved court cases are not pursued by the police.

“The deltas are downloaded to patrol cars every three minutes, so that drivers who were, say, delinquent in paying a fine or fee at 8 a.m., but have taken care of the charges at 10 a.m., will not be pulled over needlessly later in the day,” Hammad said. “People don’t want to be inconvenienced, and the Dubai Police want this system to be well-received. As the main database is updated, the patrol cars get updated.”

Before ANPR, an officer within Dubai's 15,000-member police force would would comb through lots and garages full of parked cars looking for a match against a printed list of wanted plates. This was not only inefficient, it also meant that the person of interest was not actually present, a problem overcome with an automated system that works in real time and in live traffic. A major challenge in the development of ANPR is that the system must work day and night and read a wide range of number plate styles within the seven Emirates of the UAE, which incorporate both English and Arabic lettering and different color codes.

“If you look at what we do for the Dubai Police, it’s effectively an IoT deployment,” said Humphrey. “It’s an IoT problem, because the action, forces and capacity to do license plate recognition and some of that analysis on real time video from high definition cameras is a significant CPU resource requirement.”

He said the Dubai Police are expected to expand their surveillance capabilities with version 2 of the system. It will incorporate 360-degree video surveillance embedded in the patrol car’s roof-mounted lightbar. According to HPE, the new system will deliver a more visually subtle integration of data from 12 cameras, along with the ability to read license plates up to three lanes to the right or left of a patrol car at high traffic speeds. It also will integrate car make and model data to ensure that license plates have not been attached to the wrong vehicles.

As ANPR becomes more sophisticated it will take on more difficult surveillance challenges, such as anti-terrorism. Humphrey said this is expected to happen in Dubai, and that ANPR already is being used in this role in Afghanistan and other countries.

“If the database says the registration number belongs to a Chevrolet but you’ve just seen that the registration number is on a Ford, then you’ve got a suspicious vehicle,” Humphrey said. “Why is the wrong plate on the wrong car?   This is an important capability for identifying criminal and terrorist activities with stolen vehicles.”

Asked about “Big Brother” implications ANPR, Humphrey said the system does not violate basic civil liberties when deployed to identify suspected law violators, as opposed to surveilling the public at large. He said the system has been deployed in the United States. “Constitutionally, the bigger issue is ‘watching for the sake of watching,’ as opposed to looking (for suspected law violators),” he said.