License Plates Drive Video Improvements
VIDEO RECOGNITION software continues to improve, edging us closer to intelligent analysis of images that doesn’t require human intervention. But real-world limitations still intrude, as is evident with software that recognizes license plates.
Optical character recognition, best known for identifying cars that pay tolls electronically via the use of transponders, is also used for access control, such as identifying motorists by their plate number and lifting a gate or door so that motorists can enter a parking area. This application, though, is still far from technologically perfect, say experts.
The technology works close to 100 percent in the lab, but plates are easy to read there, and they are compared to an enrolled database of plate numbers, says Lee J. Nelson, president of Electro-Optical Technologies, which helps end users select solutions from the more than 70 plate-recognition systems on the market. Plate recognition is seeing incremental improvement, but it is “far from mature,” he says.
Many factors push that 100 percent success rate down in real applications, including plate covers and brackets, sun and headlights, precipitation, odd fonts, and dirty tags. One vendor, SAIC, promises customers an accuracy
One vendor promises an accuracy rate of 80 to 90 percent of “machine-readable” plates, meaning “unobstructed, legally mounted plates,” says Bill Rapp, business development manager in SAIC’s Security and Transportation Technology Business Unit.
Nelson says that these products are best used for applications in which relatively few users are enrolled, such as at an entrance to a gated community or at the access point to an executive parking area.
Vendors are working hard to overcome these limitations. Extreme CCTV has released a product that can read plates even with intense headlights fixed on the camera, and Nelson says big advances have been made in recognizing plates in shadow and plates that are rotated or skewed.
At Hi-Tech Solutions in Lexington, Massachusetts, technologists are focusing on achieving “better contrast and improving image quality as well as robustness to make the recognition less dependent on type of plate,” says President Donald B. Brick. Brick’s company is also working on ways to determine a plate’s state or territory, since state names are often difficult to read.
Some problems are proving intractable, however. Plates with red characters, such as those used in Illinois, still give systems trouble due to the spectral sensitivity of cameras. Equally confounding are both standard-issue and vanity tags that are festooned with graphics. Florida, for example, decorates its plates with an orange. The orange has a leaf that “sometimes touches a character and in doing so creates a new character in a sense,” says SAIC’s Rapp.
The manatee on some Florida plates causes similar problems. And the profusion of different types of plates, even within a single state, thwarts progress in recognition. Nelson says that Virginia alone has almost 400 different types of license plates.
The technology would get a big boost if license plates became standardized in a machine-readable format, says Nelson. Yet that’s unlikely to happen given that license plates are a state issue and that offering a variety of plates is both a source of government revenue and an expression of state individuality. In fact, Nelson says he was commissioned to write a white paper on machine-readability for government officials in Florida. “They ignored it,” he says.