ODU eyes real-time flood alerts
Researchers to use light detection, ranging in creation of 3D map of road surfaces
NORFOLK — Hampton Roads motorists know the conundrum well: You turn the corner — after anywhere from a few minutes to a few hours of rain — and there’s standing water blocking your path.
But how deep is it? Can you drive through or will you have to find another path?
That’s a problem researchers at Old Dominion University are working to fix, armed with a new $1.5 million grant from the National Science Foundation.
The aim is to develop a system that can — on its own, through artificial intelligence — detect spots that have flooded and send alerts to drivers notifying them of problems on their route.
Khan Iftekharuddin, associate dean of the Batten College of Engineering and Technology and one of the project’s leads, said he first had the idea a few years ago when he was headed home from the university and got stuck behind a tractor-trailer struggling in floodwater.
“We tried all different ways to get
out of it but there was no way out,” he said. “It was a really challenging situation.”
It got him thinking that there should be a way to get that information ahead of time.
Fast forward several years and Iftekharuddin and many others have been working to get the project off the ground. A small Department of Transportation grant allowed some work, but the NSF funding will open it up much further.
The first step is gathering data — lots of it. For machine learning to work, it needs a great deal of information on which to base its predictions.
The researchers plan to use mostly surveillance video images from public agencies, as well as some sensors on the ground, said Mecit Cetin, an engineering professor and director of the school’s Transportation Research Institute. They’ve been in talks with the city of Norfolk and others to get access to existing resources.
The algorithm they plan to build will detect floodwaters in real time — as well as assessing how deep it is.
That’s one of the most challenging parts, Cetin said. You can’t tell from a video how deep water is, so the algorithm they develop will have to find a way to do that.
The researchers will use lidar, short for light detection and ranging, to help create a three-dimensional look at the road surfaces, Cetin said.
Camera images can show floodwaters’ edges. The lidar-fueled 3D map will have the road’s parameters. Combining the two, Cetin said, can help determine how deep the water could be.
After gathering enough data, researchers turn to developing the predictive system itself. That means simulating realistic scenarios and tinkering with modeling.
One big question that remains: once the team has a working system, what’s the best way to communicate that information to commuters?
We are overloaded with information in our daily lives, Iftekharuddin said. So Jing Chen, an assistant psychology professor, will conduct experiments with volunteers in a driving simulator, looking for the most effective way to communicate hazards to drivers.
Other team members, including at the University of Virginia, are studying the problem from the angle of sea level rise and tidal flooding, using physics-based modeling that can later be merged with the engineering side, Iftekharuddin said.
The idea is not to build a new app but eventually integrate the system into existing ones.
A few years ago, Norfolk teamed up with the navigational app Waze to do something similar. Officials started collecting data from Waze users in the hope the information could be used to start predicting flooding. The city could then alert people that certain roads would be flooded during the next morning’s commute, for example.
Cetin said that project is ongoing, but requires people to manually enter flooding instances and push out alerts as it happens. The ODU team’s system would automatically recognize issues and notify people in real time.
Though researchers all over are studying flooding, this particular approach is novel, they said. (Iftekharuddin noted it must be; the NSF doesn’t fund repeats.)
We are “going to get a very good prediction on when and where you’re going to see floodwater on the roadway surface,” Cetin said. “And that is going to be very useful information, because as of now, there isn’t a cost-effective, scalable system to collect and make use of that information.”