Background:
In 2018, Perceptics partnered with an integrator to deploy our Optical Character Recognition (OCR) software at a large U.S.tolling authority. Every year, the Authority sees over 160m vehicles of all shapes and sizes, from all over North America, on its roadways. This results in one of the most diverse license plate distributions on the continent. Collecting tolls from commercial vehicles and out-of-state travelers is crucial to the operation of the Authority as approximately 40% of the toll revenues collected are from the commercial sector and 40% are from visitors. In addition to a large out-of-state plate population, approximately 3% of plates seen are issued by Native American tribes.
The Challenge:
The legacy OCR software was having difficulty meeting the required specifications to produce a yield of at least 90% and an error rate of no more than 0.5% on all readable plates except for temporary or foreign plates. The software had difficulty reading damaged and dirty plates, which are very common on commercial vehicles, and Native American plates with intricate designs and multiple stacked characters in different locations on the plate.
The Solution:
In 2021, Perceptics rolled out its new OCR software, the 82ALPR Software, and our partner started installing it for the Authority, one turnpike at a time. Our team set to work right away to ensure the system was tuned to the most optimal level. All team members who are responsible for tuning are co-located at our headquarters, making the process of updating software configurations quick and seamless. The Authority received updated configurations for each turnpike within a few days of the initial rollout. Updating the software within days as opposed to months keeps the Authority operating efficiently and reduces the time, hassle, and cost it takes to review and retroactively correct their data. The 82ALPR Software uses artificial intelligence and deep learning algorithms to improve the ability to locate the license plate in the image and identify the state and plate type by looking at the entire plate – including the context, background graphics and symbols – instead of attempting to recognize individual symbols or characters. This new approach allows the software to correctly identify more plates that are dirty or damaged and is more flexible in reading images taken from cameras with imperfect setups. The industry estimates that approximately 7% of images of license plates are considered unreadable¹, many of which are due to the condition of the plate and image quality.
The Results:
Currently, the Perceptics 82ALPR Software is deployed system wide with the Authority and is achieving a yield rate of 91% at an error rate of 0.40% in identifying the license plate number, state, and type.
Character accuracy, state accuracy and the ability to find the plate in different parts of the image have all greatly improved. This includes better results on out-of-state vehicles, commercial vehicles, and vehicles with Tribal plates. To date, the software has correctly recognized over 200 different plate types on the turnpike with the capacity to read hundreds more.
The robustness of the 82ALPR Software allows the Authority to collect more revenue more efficiently by correctly identifying vehicles with license plates that the legacy OCR had difficulty reading. This leads to savings in the manual review process, fewer incorrect invoices being printed and mailed, a reduction in customer service calls, and an overall much better experience for the customer.
Utilizing Perceptics’ products and services can increase efficiency and revenue, while decreasing operating costs, for any tolling authority, manual review provider, or customer service operation. Interested in learning how the Perceptics 82ALPR can help you read license plates? Connect with us today to start reaping the benefits of our 40 years of image processing experience.
¹IBTTA Lost Revenue Task Force (2021, October 10-12) Lost Revenue Task Force (LRTF) Presidential Initiative Membership Briefing [Conference presentation]. IBTTA Annual Meeting & Exhibition, Anaheim, CA, United Stated.
