Startup Green Lights AI Analytics to Improve Traffic, Pedestrian Safety

For all the attention devoted to self-driving cars, there’s another, often-overlooked, aspect to transportation efficiency and safety: smarter roads.

Derq, a startup operating out of Detroit and Dubai, has developed an AI system that can be installed on intersections and highways. Its AI edge appliance uses NVIDIA GPUs to process video and other data from cameras and radars to predict crashes before they happen and to warn connected road users. It can also understand roadways better, with applications ranging from accurate traffic counts to predicting spots on roads that are most prone to crashes.

Derq CEO and co-founder Georges Aoude says his fascination with automotive safety systems stretches back to long weekend drives with his family to visit relatives. He’d wonder why these fast-moving hunks of metal didn’t collide more often.

Time revealed to him that vehicles frequently do, sometimes to deadly effect. In fact, 1.35 million people around the globe perish in auto accidents every year, and millions more are seriously injured.

“Many of our team members have been touched by deadly road crashes,” said Aoude, who himself has lost two relatives on the roads. “This only makes us more determined to get our technology out there to make roadways safer for all.”

As a scholar at MIT, Aoude worked on autonomous satellites and then drone safety, before doing his Ph.D. work on autonomous vehicles. During his graduate work, he began to envision smart cities that work in tandem with autonomous vehicles, and the seed for Derq was planted. Along the way, he received a patent for AI systems that can predict dangerous behaviors.

Dissecting an Intersection

For its initial use case, Derq zeroed in on intersection and pedestrian safety. It chose to test its technology at a busy downtown street crossing considered one of Detroit’s most dangerous. The intersection had no cameras or radars, so Derq procured those and began its monitoring work.

For the past three years, it’s been capturing and training footage of tens of thousands of vehicles and road users a day, 24/7, at that intersection. In addition to predicted red-light violations and dangerous pedestrian movements, the company has been using that data to refine its AI models. As it monitors more intersections and roadways, Derq is constantly expanding its models. All data is anonymized, and no personal information from road users is ever collected or stored.

Those models can determine which actions have the potential to cause crashes and which don’t, and Derq strives for over 95 percent accuracy. Incidents considered high risk are uploaded to Derq’s GPU cloud instance for further analysis and documentation. Eventually, the system will be able to notify connected and autonomous cars and warn them of impending dangers.

“If the driver detects someone is running a red light, that’s too late,” Aoude said. “If you tell the driver two seconds in advance, they now have the chance to react and avoid the collision.”

The Detroit test demonstrated that the system was able to identify potential crashes effectively. Having seen the technology, the Michigan Department of Transportation has expanded its engagement with Derq, which was recently awarded a federal project to deploy its technology at 65 key intersections along 25 miles of connected roads in the Motor City.

Derq has also been deploying its technology in Dublin, Ohio, a suburb of Columbus, and in Las Vegas, and is running a pilot with California Department of Transportation. It’s in discussions with cities in Florida and Texas, as well as in Canada. Further afield, it’s deployed its systems in Dubai and is working with the road and transport agency there on expanding that deployment as well.

Choosing the Right GPU

Derq’s edge units are equipped with NVIDIA GPUs and hardware acceleration that enable them to process large amounts of internet of things data in real time. It’s developing for a variety of GPUs, including the NVIDIA RTX, T4, and Jetson Nano, AGX Xavier and TX2.

But GPUs are just one part of Derq’s relationship with NVIDIA. As a member of the NVIDIA Inception program for AI startups, Aoude said Derq has had an Inception team visit the company’s Dubai office to provide support such as how to optimize their use of GPUs. They’re also in the process of being incorporated into the NVIDIA Metropolis smart spaces AI platform, which Aoude said “will be a great platform to help us scale.”

The company has focused its product on two main unit sizes. A basic unit runs a single application — think of a box that’s deployed at a pedestrian crosswalk — a perfect pairing for a Jetson AGX Xavier. More elaborate boxes, which run more apps on more camera streams at complex and busy intersections, will rely on the more powerful NVIDIA T4 or RTX GPUs.

Derq will also provide traffic planners and engineers with valuable statistics and intelligence, such as real-time graphs, incident notifications and heatmaps, that can help them assess and improve road safety. The system also can provide forensics for government agencies and insurance companies investigating crashes, an area that the company is exploring with insurance providers.

Going forward, Aoude said Derq plans to scale quickly through partnerships with cities, smart infrastructure, and autonomous vehicle firms so that it can get its technology deployed as widely as possible before autonomous vehicles hit the roadways in large numbers.

“We can’t do it on our own,” he said. “We need forward-thinking cities and collaborative partners, and get ready to scale deployments together, to achieve vision zero – eliminating all road fatalities.”