Engineering Scalable Technology Through Industry Change

Peter Crary, Deputy Director of Technology Services, Texas Association of Counties

Peter Crary, Deputy Director of Technology Services, Texas Association of Counties

For over a decade I worked in the controversial and often misunderstood world of automated license plate recognition (ALPR) as a vendor for one of the world leaders in that industry. In that time, I watched the technology grow, evolve and adapt to changing technological and political climates. By the time I hung my hat up to join county government, I was leading an international team of software, mechanical, firmware and optical engineers to design and produce technology still used all over the world.

The technology is simple in concept and complicated in its design. Although the later expansion of artificial intelligence (AI) transformed that industry. Prior to AI, the concept was to have a camera system find and take a multiple pictures of a license plate for the purpose of trying to read it digitally. This information could then be used in the early 2000sIn the late 1990s, the technology was limited to bulky camera and poles (called fixed cameras), but when ALPR was brought to the US a short time later in the early 2000s, the cameras and computer processors were separated allowing the cameras to be mounted on the police cars while the processor was installed in the truck (this configuration was called mobile cameras). Two things made this technology special. Multiple cameras could be mounted on a police car allowing the vehicle to detect multiple vehicles simultaneously. The mobile data computer or MDC the in-car computer, could trigger an alert when a wanted plate was detected. This allowed law enforcement the ability to passively look for anything from stolen cars to late registration and act. It also had a profound impact on investigations that later shaped the entire ALPR industry.

Early adopters of the technology, such as Cincinnati PD, San Diego County, Los Angeles PD and many others saw the potential of the technology as a tool for solving crimes. It was no longer just a tactical tool for finding a wanted vehicle in the moment. It was now a tool for identifying vehicles near crimes, finding patterns in criminal movement, identifying criminal associates and rapid crime response.

“Two things made this technology special. Multiple cameras could be mounted on a police car allowing the vehicle to detect multiple vehicles simultaneously. The mobile data computer or MDC the in-car computer, could trigger an alert when a wanted plate was detected.”

One of the more notable instances of this was in 2015 when a news reporter and cameraman were shot on live tv in Moneta, Virginia. The suspect fled in a car, but police quickly entered the vehicle plate into their ALPR system. The system identified that the vehicle had recently passed one of the cameras and giving officers an idea of where he was and the direction he was traveling. Within 3 hours of the attack, the suspect was caught where he took his own life.

While the camera and the optical character recognition (OCR) technology did not evolve too much, at least not until the AI boom, the investigation software rapidly changed. For my role as Director of Research and Development, this posed challenges in managing, developing and distributing a rapidly growing list of requirements.

Prior to this, the typical engineering cycle was a waterfall methodology to ensure the software, firmware and hardware were all aligned. This model had to change. By the early 2010s, we switched our software to an agile methodology to address the rapidly changing requirements. The software team itself, consisting of five developers, one database administrator and two quality analysts, went through the typical forming, storming, norming, etc. to where we were able to rapidly develop features for our wide customer base (approximately 1000 police agencies over North and South America, Europe and Australia/New Zealand). Over the next decade we rolled out many features including enhancements that could identify common plates and multiple crime scenes due to a serial bank robber, a multicar caravan algorithm that could help identify drug runners and gang affiliates, or a proximity detection of know sex offenders within a radius of a school or playground.

Then came AI. AI allowed ALPR to do more with little. Cameras no longer needed to be large, bulking systems because almost any HD camera could be paired with an AI engine to perform the OCR and vehicle identification. ALPR could now identify a vehicle that did not have a license plate based on the unique characteristics of the vehicle such as color, make, rear spoiler and even bumper stickers.

On the development side, this was very exciting, but it also caused a huge shift in the industry. Vendors that were already poised around AI were suddenly becoming the industry leaders. The original industry leaders with traditional cameras and OCR had to adapt. Many of them were bought out or shifted away from law enforcement to other industries like open road tolling. The subtle shift to software before was now a major shift towards software and AI. My experience in this industry during its time of significant change taught me a lot around adaption, open to a changing landscape and willingness to learn.

AI

Agile