FNINDING A PURPOSE
The most crucial part of project planning is determining the project's purpose. An agency that wants to monitor road conditions may not need crystal-clear images for facial identification or LPR.
Some policies drive trafficmonitoring implementations. For London, its congestion charge started as a policy to deter people from driving. “The charge is enforced using LPR,” Cartwright said. “For cost and liability, London needed to ensure all the entry points to the charging zone are monitored.”
Other citywide policies aim for better living by managing congestion. “Eliminating traffic from specific zones in the city improves livability,” Hobbelink said. “Monitoring and management of restricted vehicle access is, therefore, a hot topic.” Traffic systems are becoming more informative as well. “Throughout the world, we see a shift away from monitoring to proactive traffic management,” Swaminathan said.
“This is driving changes in the information system. You can see the emphasis moving from traditional dependence on front-end equipment, such as cameras, to software and analytics. Operators can use predictive analytics to anticipate a potential incident on a road that is getting increasingly congested. They can then position emergency vehicles on roads, rather than wait and react to an incident.”
IP?
The next stage is to determine what technology is best suited for a particular application. IP video surveillance has tremendous benefits, but before procurement, it is essential to understand what its limitations are. Megapixel cameras offer impressive amounts of detail, but require a supporting cast to function in an outdoor environment. “High resolution is almost not desired; it's seen as a luxury,” Cartwright said.
Nighttime imaging is also an issue, as megapixel cameras require IR illuminators. One illuminator may suffice for one lane of traffic, but not across four lanes. More illumination will help, but overheating may cause the camera to crash. Operators will have to weigh between seeing detail and streaming video. While there are newer cameras that can cover up to two lanes with a single 810-nanometer illuminator, these are highly application-specific.
Extra lighting may simply be unavailable on highways. “With the low-light issues, none of the available megapixel cameras will do,” de Konink said. “Additionally, if you mount a camera on a pole 6 to 12 meters high, there will be wind. You will have to make it a very stable pole, but that is very expensive.”
Dynamic range also remains an issue, giving rise to noisier images and increased bit rates, said Martin Yang, Director of R&D for Sunell Technology. CMOS sensors are getting better at handling more pixels, but still have a way to go before they catch up with CCD low-light performance.
HD is a growing trend for traffic applications, but only for specific locations such as toll stations or entrance ramps. IP is seeing added adoption in traffic, but not every camera on the road should be a megapixel one.
General wide-area monitoring still depends on SD cameras for real-time transmission.
LPR
Traffic-monitoring checkpoints frequently deploy LPR as well. The world electronic toll collection (ETC) systems market is the fastest growing product segment, displaying a CAGR of about 12.77 percent until 2015, according to a Global Industry Analysts study. 
For identification, HD is hard to beat. Key places that deploy HD cameras and LPR also use PTZ cameras, said Cheng Yu, PM at ZTE Netview Technology. “In a typical traffic application, a mixture of HD and SD will be deployed to lower the cost.”
However, real-life LPR performance is not 100-percent accurate and takes up significant processing power and bandwidth. For operators, it can be hard to use as well. Some solutions try to free up bandwidth by compressing the image before running LPR algorithms, but the loss of image detail reduces accuracy and defeats deployment purposes. Edge processing will make this better, but changing conditions make LPR easier said than done.
License plates have multiple formats, making it hard to recognize them all. “In India, some plates write the numbers on one line, while others are written on two lines,” said Zaheer Ali, Director of Oriole Electronics. “Some are in plastic or metal; some are not in English but in Hindi. It's very hard to perform character recognition. LPR has never worked in India for the current format.”
MULTIMEGAPIXEL AND BEYOND
Traffic monitoring is a really demanding situation where operators need to be able to identify vehicles, objects or even persons in a scene, said Nafis Jasmani, Sales Manager for ASEAN at Axis Communications.
“This higher resolution enables the operator to choose either to zoom in or to maintain a wider field of view without compromising the ability to see and discover fine details in the image.”
However, escalating megapixel counts need to be weighed against frame rates . “ I f you t ake 20 megapixels and you zoom 10 times after the fact, the camera could look for license plates, faces or through the windshield to faces,” O'Malley said. “However, if you reduce frame rates while cars travel at 160 km/hr to 4 fps or the IR illuminator is shining at the wrong place, no matter how great the lens or camera is, you can't use that data.” Megapixel is useful in specific traffic applications, but is not practical for every camera system-wide.