Learn More
REKOR News, Articles and Studies on ALPR and Videos

ARTICLES

What ALPR Has Learned from Artificial Intelligence and Machine Learning

by Matt Hill, Founder of OpenALPR

Artificial Intelligence (AI) has enabled applications that, only a few years ago, seemed impossible.  Active development of this emerging technology is yielding self-driving cars, facial recognition, augmented reality, and much more.  OpenALPR uses AI in its License Plate Recognition (LPR) engine to achieve superior accuracy compared to legacy products.

What is AI

The combination of massive volumes of data, large-scale computer resources, and specialized software algorithms called neural networks, combine to enable computer programs to learn to make remarkably accurate predictions from imperfect inputs. 

he breakthrough that began the recent AI boom was in 2012 when Alex Krizhevsky presented a solution, named “AlexNet”, to the ImageNet research challenge. AlexNet utilized a novel convolutional neural network architecture to classify images.  While other teams were researching solutions that provided a 2-4% incremental improvement over the previous year’s best result, AlexNet was roughly 40% better than the second-place team.  It was immediately apparent that this was a superior approach, meriting further research.  In the following year, and every year since, all of the top scoring competitors use some variant of neural networks in their submissions.

This disruptive technology has had a similar effect across virtually all software fields where it has been successfully adapted.  Software that was once state-of-the-art is becoming obsolete as AI solutions perform significantly better.  AI has also found utility in automating tasks that previously required human intuition, such as analyzing radiological studies and logistical planning.


How
is OpenALPR Different?

The short answer is that is significantly more accurate and it can detect vehicle make, model and color using low-cost consumer-grade video cameras.

OpenALPR began as an open source project in 2013.  The software originally employed traditional computer vision techniques to recognize license plates in images and videos.  This initial approach was similar to how other commercial LPR products currently operate.  The effectiveness of traditional computer vision is limited because it can only achieve a moderate level of accuracy, usually about 80%, before reaching a plateau.  OpenALPR commercial software uses an exclusively-AI approach which increases accuracy significantly.  Further, as the AI training dataset becomes larger and the algorithms are fine-tuned, OpenALPR’s accuracy continues to improve.

OpenALPR publishes verifiable benchmarks that demonstrate its software’s accuracy.  

The use of AI also allows OpenALPR to detect information such as the make, model, and color of the vehicle.  This detection is based solely on surveillance video data and does not require the use of a separate vehicle registration or government database.  And with fixed-based camera’s, OpenALPR can also detect a vehicle’s travel direction. This type of vehicle information becomes crucial to law enforcement when eye witnesses can recall the vehicle information, but not the license plate number.

Why Does it Matter?

Traditional LPR has been available for more than a decade, but it required specialized, and often proprietary, camera equipment to simplify visual data so that software could pick out the license plate data.  This equipment is more expensive than consumer-grade, mass-market video cameras that can be used for a variety of purposes.  Because OpenALPR’s software is AI-based, it can effectively us image data from low-cost, commodity cameras while also being more accurate – a better and economical solution.

LPR is widely used by law enforcement, for security, and by the vehicle repossession industry.  LPR  can also be used for parking lot management, car wash loyalty programs, fast-food drive-through lanes, logistics tracking, automated ticketing, etc.  In the past, LPR was not considered for these purposes because it was not accurate enough.  By increasing recognition accuracy from less than 80% to 99%, OpenALPR’s software is enabling transformative automation solutions to be viable.

OpenALPR provides a solution that is significantly more accurate and cost-effective than other products on the market.  Not only does this make it more compelling to deploy more LPR cameras for law enforcement, security and commercial purposes, it also opens new markets for the technology.  In the near future, we expect to see a dramatic increase in the number of license plate recognition devices installed across the world. 

Consider briefly explaining in a call out, a neural network.  I found the following on Wiki “ A neural network is framework for many different machine learning algorithms to work together and process complex data inputs.[3] Such systems “learn” to perform tasks by considering examples, generally without being programmed with any task-specific rules.” We could add a similar callout box to briefly describe “machine learning”

VIDEOS

'Move Over' Law News Stories and Media

WHITE PAPERS & CASE STUDIES

Case Study: Uninsured Motorists

Case Study: After Hours Tresspassing