Amir Hever was driving into a government facility a few years ago when he discovered a huge flaw in their security process. As he approached the entrance gate, a security guard dropped to his knees to look underneath his vehicle.
"When he stood up, I asked him what he was looking for," said Hever, CEO and co-founder of computer vision startup UVeye. "The security guard answered honestly that he was looking for threats but actually couldn't see anything. That's when I realized that something wasn't working right."
Hever assembled a team, and began researching the problem and potential solutions. Thus was born in 2016 UVeye, which has since built an under-vehicle inspection system that uses deep learning to bridge the security gap.
Much of the New York-based company's work centered on grasping the vast variety of vehicle undercarriages, not to mention the changes they undergo after thousands of miles on the road. What Hever and his team learned that it's not easy to identifying anomalies in vehicle undercarriages.
"We didn't know what we're looking for as there is no standard of what a threat would look like in the undercarriage," said Hever. "Moreover, threats are usually concealed."
More Than Schematics Needed
UVeye quickly learned that schematics provided by vehicle manufacturers aren't enough because, after thousands of miles of road time, undercarriages don't look like they did when they came off the assembly line. The answer was to develop an algorithm for unsupervised learning that would make it possible to spot threats -- no matter how well concealed or the condition of the vehicle's undercarriage.
The company rented hundreds of vehicles in various conditions and scanned their undercarriages, generating both 2D images and 3D models. That data was fed into its deep learning model, which maps the location of all the parts (segmentation) and then analyzes each segment separately and looks for anomalies.
This allows it to detect any alterations or anomalies to those parts, or the presence of foreign objects as small as USB drives. It can also tell whether a chunk of snow or mud looks natural, or if it might be a disguise used to conceal something.
UVeye uses workstations running multiple NVIDIA GeForce GTX 1080 GPUs to train its models. It turns to cloud-based GPUs running on Amazon Web Services or Microsoft Azure to train beyond its workstations' capabilities, or to speed up the process further.
Hever said the use of GPUs, as well as the CUDA parallel computing model, significantly sped up the company's training and development processes, as well as the system's ability to generate results.
UVeye's first line of products enables customers to automatically scan, detect and identify anomalies, modifications or foreign objects in the undercarriage of any vehicle. The company has already installed its system -- packaged as a piece of hardware that sits in the ground, scanning vehicles that pass over it -- at more than 30 sites worldwide. This has provided abundant test data verifying the system's effectiveness.
"Our machine learning algorithm detects anomalies in any vehicle whilst in motion, within three seconds," said Hever. "GPUs are making it possible."
Inspection as a Service
Today, UVeye is revolutionizing vehicle inspection for the automotive industry. Additional applications for its system focus on security, with homeland security representing a robust market for the company's technology.
"The need for an automatic external inspection system for vehicles that can detect anomalies, changes and dents, and also track changes over time, is huge," Hever said.
The company has also leveraged its algorithms to analyze other parts of vehicles besides undercarriages and to inspect any vehicle from all sides.
"UVeye's 360-degree system can detect vehicle leaks, wear and tear, and a wide variety of mechanical problems or damages," Hever said.
From auto sales and rentals to fleet management and maintenance, Hever sees infinite opportunities for his company's Inspection-as-a-Service model to ensure safe and reliable operation of vehicles.
Said Hever, "We are going to change the way that people and organizations inspect their cars."