The intersection of Jefferson Avenue and Randolph Street in downtown Detroit is often cited as one of the most dangerous in the metro area. But that could change in the near future, thanks to a recently completed pilot project that aims to nip accidents in the bud by alerting drivers in advance to dangerous situations.

That project was spearheaded by Dubai- and Detroit-based artificial intelligence company Derq, in partnership with the city of Detroit, the Michigan Department of Transportation (MDOT), and the Michigan Economic Development Corporation. Last spring, Derq installed over 10 sensors, including video cameras, thermal cameras, and radar sensors, at Jefferson and Randolph to monitor vehicle and pedestrian behavior. Derq, a PlanetM Landing Zone member, used a PlanetM grant to fund the project.

Dr. Georges Aoude, Derq co-founder and CEO, says Derq chose the intersection for its extreme complexity. It connects two major highways and the Detroit-Windsor Tunnel, and also has multiple major pedestrian crossings. What’s more, Aoude says, “the rules are different depending on how you’re approaching it.”

“You have two right turns on red, one unprotected left turn, one mandatory right turn, two forbidden left turns, (and) one forbidden right turn,” he says. “So you see a lot of road users violating those rules on a daily basis.”

Aoude and his team set out both to study how people were using the road, and to begin to predict that behavior. Derq installed industry-standard sensors at the intersection, but combined the data the sensors recorded with Derq’s own machine-learning algorithms.

“That way we’re not able to only have these sensors see, but they can also think, detect, and predict in real time those dangerous situations,” Aoude says.

The end result is a system that has successfully learned to anticipate a car running a red light, or a pedestrian stepping out in front of traffic, two or more seconds in advance.

“Those few seconds can make a big difference between potentially colliding with a person and coming to a safe stop,” Aoude says.

The system has yet to prevent any accidents, but its potential will likely be actualized very soon as vehicle-to-infrastructure communications technology comes into play. A sensor that detects a vehicle or person about to cross against oncoming traffic could then send an alert to any connected vehicles at risk of a collision. Those drivers would then receive an advance warning on their dashboards to slow down or stop. Autonomous vehicles could receive a signal to slow down or stop on their own.

The project also has broader implications for addressing the underlying issues that cause accidents in the first place. Aoude says the sensors collect information on overall traffic flow and can therefore identify dangerous trends – for example, the way the timing of a traffic signal may make drivers more likely to run a red light at a given time of day. Aoude says road operators will be able to reduce accidents on a case-by-case, short-term basis, but “in the long term they can make changes that will improve the traffic flow, the schedules, (and) the rules.”

Aoude says the Jefferson-Randolph project was particularly helpful for Derq because it was one of the most complex intersections the company had worked on at the time. The company’s previous installations had been in Dubai, so Detroit’s very different weather patterns provided an additional layer of challenge.

“We had to adapt and learn how to handle the complexity,” he says.

As a result, Derq has taken on projects of similar complexity in additional markets. It now has an installation in Vienna, and Aoude says a project in Ohio will likely be announced soon. Here in Detroit, Derq’s one-year pilot period is over. But Aoude says the company is in discussions with MDOT about potentially adding more services to the Jefferson-Randolph intersection, adding more sensors in the immediate area, and doing projects in other Michigan cities.

“This is a continuously ongoing process,” he says. “As we collect information, as we keep training our algorithms, we’re learning new things.”