Cutting-edge software can extract high-quality ALPR data from any CCTV camera feed. Police departments are already using such systems for surveillance and crime prevention, but how useful can they be for traffic enforcement? Jan Stojaspal investigates
The police department in Mobile, Alabama, makes dozens of arrests a month based on information from an intelligent camera surveillance system that is able to read license plates and flag those that belong to a stolen vehicle or a known perpetrator of a crime. Yet it does not own a single surveillance camera, nor does it intend to purchase any.
“In Mobile it is not really the place of the police department to go out and buy thousands of cameras,” says Commander Kevin R Levy, technology and cyber-intelligence section, Mobile Police Department. “There isn’t money for that, and it is not the role a police department is typically taking. What we are doing is harnessing relationships with people who already have cameras.”
Mobile’s use of camera surveillance technology that can read vehicle registration plates is not new. Police departments all over the USA and elsewhere have been using automatic license plate recognition (ALPR) for the better part of two decades. What makes Mobile noteworthy is its reliance on an entirely camera-agnostic system, which is where ALPR appears to be heading next.
Systems that do not require a vendor-specific camera or a vendor-specific integration between the camera and the ALPR software are cheaper to assemble and operate, and can extract a much wider range of information about a vehicle than just an alphanumeric ID, thanks to cloud-based computing that harnesses artificial intelligence and machine learning.
In addition to the alphanumeric ID, the system in Mobile can also deliver the vehicle’s make and model. It captures a variety of traffic flow data, including the volume and density of passing traffic. And it makes for easy travel-time calculations. “It’s kind of like Disney World does with lines,” Levy says. “You can figure out the travel time between two cameras and say, hey, it’s taken somebody 20 minutes to get from here to here; it is going take them two hours to get out of state instead of 30 minutes.”
The traffic flow data came in handy last fall when residents of Florida evacuated through Mobile ahead of an approaching hurricane, and the police department was able to use the data to gain greater visibility on where to redirect traffic. “We have traffic cameras where you can see cars,” says Levy. “But seeing a bunch of cars and knowing how many are passing through an hour are two different things.”
According to Lieutenant Brian Hess at the real-time crime center of Westchester County Police, New York, an ALPR system that is analogous to Mobile’s helps his police department look for stolen cars and people wanted for crimes. But it also allows the local department of public works, responsible for road maintenance, to count cars.
Neither Mobile nor Westchester use their ALPR systems for traffic enforcement – like red light or speed cameras – for reasons of privacy and local traffic enforcement laws.
Although Westchester County Police has been operating traditional ALPR units, both fixed and mobile, for some years, it turned to camera-agnostic ALPR last year because it was much cheaper.
“The problem with vendor-specific systems is the expense,” Hess says. “One traditional ALPR setup, such as a fixed site on a highway, would cost us between US$75,000 and US$100,000. For the same amount I could probably deploy six camera-agnostic sites.”
Not all surveillance cameras produce video feeds that are ALPR-ready. Of the 5,400 feeds that the Mobile Police Department has access to (as part of Project Shield, a public-private partnership that gives the department’s cyber-intelligence unit access to surveillance cameras in businesses, schools, residential communities and other establishments in and around Mobile) only about 80 are good enough to be processed for license plates information. But that number may be as high as 1,000 within a year as Project Shield expands, Levy says.
Vendor in control
According to Alan A Quinn, a USA-based independent consultant who specializes in video detection, license plate recognition and CCTV surveillance, a dedicated ALPR system will outperform a camera-agnostic system because the vendor is in control of the setup, positioning of cameras and illumination.
“Normally speaking you will find that the vendors will guarantee you a percentage of captured plates and the percentage accuracy on those plates,” he says. “It’s very difficult for them to do that if there are a variety of camera technologies out there.” According to him,a camera-agnostic system is far more cost-effective, but its “accuracy level will fluctuate” in low light or different weather conditions, for example.
But this is beginning to change, according to Steve Lewis, vice president for business development at openALPR, the software-only company that powers the camera-agnostic ALPR systems for both Mobile and Westchester police forces. “Any camera can be an ALPR camera today,” he says, provided it can deliver a sharp enough image and the license plate in that image is at least 100-120 pixels wide.
“We are dispelling the myth that the industry has perpetuated for decades that you need a special – i.e. expensive – camera,” Lewis says. “Of course, if you want to capture license plates at night you will need a camera that is day/night capable and have some internal or external IR illumination.”
Lower costs mean that ALPR can spread beyond the traditional confines of law enforcement and city-wide surveillance. It’s worth noting, however, that dedicated cameras often have usability and durability advantages, being built for road use. They are weather-sealed, come with external lighting and are also built to last. Being able to take any camera off the shelf does not mean it will last through the rigors of being used 24/7 in all types of weather.
For whom the ALPR tolls
Open-road tolling is one growth area for ALPR that is based on machine learning. This is because traditional ALPR systems, which are typically based on character matching, struggle with specialty and vanity plates, and also with state indicators in the USA.
Other emerging applications, according to Lewis, include parking access, VIP recognition, tracking the efficiency of delivery in logistics, and so-called frictionless retail payments where the license plate authorizes payment in a drive-through restaurant or a gas station.
Also important to ALPR’s future is what one does with the data, particularly whether it continues to be kept in agency-specific silos or is put to wider use. In this respect, the Automated Regional Justice Information System (ARJIS) is a sign of things to come.
According to Dale Stockton, an ARJIS project coordinator, the system was created to share criminal justice data among more than 80 local, state and federal agencies in San Diego and Imperial counties in California.
“Although that sounds like common sense, and maybe on TV it is the norm, in real life it’s pretty much the exception,” Stockton says. “Of the over 15,000 law enforcement agencies in the USA, the vast majority don’t have access or at least ready access to what agencies literally next door to them are doing.”
ARJIS enables authorized law enforcement personnel to query ALPR data from a much wider geographic area than would be covered by a single agency and, at the same time, to match the data against a much wider range of criminal justice information sources, such as citations, arrests and crime reports.
And this is key when it comes to preventing or solving complex cases such as terrorist plots. “We recognize that there is great concern when it comes to privacy, and we have policy and practice to safeguard the data,” says Stockton. “However, it’s also important to point out that license plates are issued by the state and the law requires that they be displayed on every vehicle. By default they are in public view and law enforcement is empowered to observe and check those plates.”
While there is the potential to use software-only ALPR systems for daytime parking enforcement, and they will certainly help with making plate identification more accurate in open-road tolling,it’s difficult to see possible pure traffic enforcement applications for camera-agnostic ALPR in the near future. Currently software-only solutions don’t provide the level of robust reliability required for legal enforcement. Only dedicated camera vendors can guarantee that.
The reasons for this include the special build for a camera required to withstand 24/7 road use, to work at night and deliver sharp images from vehicles that are traveling in excess of 70mph (110km/h). This last point alone eliminates cameras that would work for a managed parking facility, for example.
Another important consideration is that traffic management systems are typically not sold or operated piecemeal. They are often huge commissions that require a great deal of integration and customization. “For highways and tunnels we are talking about very big tenders, so it’s not just cameras and software for automatic incident detection. Very often the tender covers the building of a tunnel, the road, plus all the technology,” says Renato Clerici, CTO, Sprinx Technologies.
But as software that makes simple cameras smarter becomes more commonplace and its uses multiply, we may begin to see much more crossover from other uses with traffic management. For example, as police departments begin to share data, there could also be benefits of sharing this data with traffic managers and vice versa. As with so much disruptive, emerging technology, the question quickly becomes not whether the technology can do it, but whether the law will allow it.
The future of incident detection
Automatic incident detection has long been reliant on software solutions. Soon improvements in AI will mean even greater functionality.
In Europe, automatic incident detection (AID), which is another way in which video analytics is used in camera-based road surveillance, is not just nice to have. It is the law. According to European Directive 2004/54/EC on minimum safety requirements for tunnels in the Trans-European Road Network, automatic incident detection is essential for tunnels over 500m (1,600ft) in length.
AID solutions are typically radar-or video-based. The former currently produce fewer false and unwanted alarms, while the latter are cheaper and “in any case cameras must be installed in the tunnel for video surveillance,” says Renato Clerici, CTO at Sprinx Technologies, an AID software and engineering company based in Italy.
Also, as video-based AID matures, it begins to offer greater flexibility, according to Andrea Sebastiani, who is responsible for electrical and technology systems on Autostrada A2 in the south of Italy. The 274-mile (441km) stretch of highway features 56 two-bore tunnels and more than 400 AID-enabled cameras.
Currently the cameras can detect events such as a stopped vehicle, the beginning of queuing or a pedestrian in the road, according to Sebastiani, who works for Anas, a government company responsible for the construction and maintenance of Italian motorways. “In the near future, I expect that systems will be able to understand vehicle behavior better, to follow and track a vehicle passing in front of different cameras and to match all this data with other technologies, thus increasing intercommunication between traffic systems,” he says.
“Moreover, I think there will be fewer multiple devices and more all-in-one solutions able to provide multiple functions,” he adds. “Last but not least, the introduction of artificial intelligence will increase performance, rates of detection, and add additional information not currently available with a standard AID system.”
