AI on the road: How automatic license plate recognition works

This article was written and published in Spanish and has been translated into English via Google Translate. Click here to read the original article.

Circulating on the Spanish roads it is possible that you have ever noticed some cameras that, located next to some information panels, overpasses or traffic lights, are dedicated to watching all those cars that pass under them.

Every time a speeding ticket reaches your hands, it is thanks to the main functionality of these cameras: automatic license plate recognition, also known for its acronyms in English ANPR (Automatic Number Plate Recognition) and ALPR (Automatic License Plate Recognition).

ANPR cameras dedicated exclusively to license plate reading are capable of recording perfectly visible snapshots of vehicles in motion (even those that drive at high speeds) thanks to their remarkable shutter speed: it is usual to find devices of this type with a shutter speed of 1 / 10,000 (that is, requiring only a ten thousandth of a second to capture the image).

This allows them to be used both to know which cars have passed through a certain geographical point (although there are also mobile ANPR cameras, prepared to be coupled to police cars). The capture of images during the night and in adverse weather situations (rain, fog, smoke, etc.) is solved by using cameras with infrared reflectors.

This technology uses machine learning to differentiate license plates from other visible elements in the image (such as traffic signs or advertising labeled in cars) and OCR (optical character recognition) technology to identify the letters and license plate numbers in question.

The next step, once the registration characters are recognized, corresponds to the comparison of the same in a database to identify the vehicle in question and the information available on it (model, list of owners, etc).

License plate detection can be performed with cameras specific to this technology (which perform the task of recognition within the device itself), but it is also common to run ANPR software to analyze the video provided by a closed circuit television.

Although more recently, with the availability of open source software prepared for this task, it is possible to use, in practice, all types of cameras, from those of an autonomous car to any webcam.

Police forces around the world use this technology both for traffic management purposes, detecting violators of road regulations (in Spain, the DGT has provided its network of ANPR cameras with an extra function: the detection of seat belts) as for surveillance and prosecution of criminals.

Toll road management companies, meanwhile, use it to control their payment; Similar uses give some service stations. Likewise, it is already used to automate access to restricted access car parks. But new uses for the ANPR constantly appear.

A story that began 43 years ago

The first technology focused on the recognition of license plates (although it was not yet automated) was developed by the Division of Scientific Improvement of the British Police in 1976. To this contributed both the increase in computing power and the improvement of the cameras of the moment.

But it did not begin to become popular until the 90s, when the software became much easier to handle and the hardware much cheaper. The collection of ANPR data for future use (that is, in order to solve crimes not yet identified) had to wait another decade. It was in 2005 when this technology allowed, for the first time, to identify and detain three murder suspects.

And what is the future of this technology in the short term? The latest studies, published just a few days ago, show that ANPR solutions are a growing field whose value will increase by 9.3% per year by 2024, to a total of 4.7 billion dollars.

Another threat to our privacy?

But the popularization of this technology is a two-sided coin: many argue that the existence of a broad infrastructure of ANPR devices should raise the same debate about the role of privacy as that caused by the massive adoption of facial recognition since, at After all, it continues to allow us to know the location and routines of citizens.

And just as the diffusion of facial recognition has motivated activists to look for all kinds of creative techniques to circumvent or manipulate it, license plate recognition begins to elicit the same kind of reactions.

On the other hand, the Electronic Frontier Foundation has also denounced the danger posed to our privacy by the negligent use of this technology: a study published in 2015 by this US organization found dozens of ANPR devices exposed on the Internet because of poor configurations of safety.

Three years later, and with little effort, the online media Techcrunch found more than 150, including cameras used by California or Washington state police.

This article was written and published in Spanish and has been translated into English via Google Translate. Click here to read the original article.

Join us November 12-15 for the Property Portal Watch Conference Madrid 2019.