License-Plate Recognition via Artificial Intelligence

This post is also available in: עברית (Hebrew)

Law-enforcement agencies often use automated license-plate recognition (ALPR) technologies to identify the location of vehicles associated with criminal activity. The readers can be mounted not only in police cars but also on utility poles, streetlights, within traffic cones and digital speed display signs. Once a vehicle’s plate is photographed, and the date, time, and location are recorded, an algorithm checks it against a database of cars that police is looking for.

The technology uses optical character recognition (OCR) on images to read vehicle registration plates to create vehicle location data. However, the specialized equipment is expensive, this resulting in license-plate readers being deployed in a limited number of vehicles.

In a move that would reduce the cost of ALPR and address ethics concerns that have been raised about the technology, Axon will integrate ALPR powered by artificial intelligence into its next generation of in-car video cameras. 

Instead of using infrared technology, the company will leverage the wide-angle video feed and AI processing power in the Axon Fleet 3 in-car video system — expected to be in the market late next year — to identify license-plate information at a 90% reduction in cost, even when a regular ALPR subscription fee is included.

“Fundamentally, what we’ve done is we’re using AI to process video streams off of a standard, color in-car camera,” Rick Smith, CEO and founder, told urgentcomm.com. “What that allows us to do is dramatically reduce the cost, because an in-car camera is around a $1,000 investment, as opposed to a $20,000 investment. “We’re able to now do license-plate reading using the stream from these in-car cameras, where it is effectively just software running on an in-car camera, rather than having to buy a whole separate system.”

“We’re actually running the AI in the camera itself,” he said. “With our existing cameras, you would have to push the video stream up to the cloud, and that would consume a ton of data.

“With the new camera, we’re putting a pretty beefy AI-capable processor in the camera itself, so it’s doing all of the image processing locally on the camera.”

The comparison of new data to a database of license-plate numbers of interest can be achieved through two approaches, he explains. “The camera will do the basic processing, and it will determine the license-plate number and state. We can compare that to data that can be stored in encrypted form. So, you can have hot lists that can be pushed down to the camera, or you can push that data up to the cloud, but what we’re really pushing then is just text data — license-plate numbers and states — which is a very light data lift, compared to pushing a whole video stream over LTE up to the cloud for processing.”

Axon is trying to address privacy and ethics concerns by vowing not to resell ALPR data commercially and developing a “heat map” functionality depicting ALPR usage that can be shared with the public, Smith said. In addition, by dramatically lowering the cost of ALPR, the technology could be deployed in most law-enforcement vehicles, so there is less concern that only certain segments of the community are being targeted—a concern voiced by some civil-liberties organizations, he said.