For nearly two years, a professor from Concordia University has been working with the City of Terrebonne to create flood simulation software. A tool that will make it possible to prepare without risk for overflows, and that emergency service employees are “eager to test”, according to the director of information technologies of the City, Rémi Asselin.
“At the start, the project was first to have equipment that would allow us to measure different elements on the river: the water level, the level of snow that has fallen, or the speed of the currents, recalls Mr. Asselin. Subsequently, the project evolved into the development of a predictive model and simulation of floods. »
Taking the appearance of a video game, the program makes it possible to reproduce the territory of the city in 3D and to “change the aspects of the terrain”, explains for his part Charalambos Poullis, associate professor of the Department of Computer Science and Software Engineering at Concordia University. Users can, for example, place sandbags around residences near the Rivière des Mille Îles, which are particularly at risk, and see how these new elements would modify the flow of water.
Using historical data from Environment Canada, the software can simulate past events, or use statistics to predict the extent of future floods. All this “at zero cost”, rejoices Professor Poullis. “There is no danger of breaking anything, you reset and you start again! That’s the beauty of it. »
The project, which is still in the development stage, will be “especially useful in the spring, when the snow will start to melt, then the level of the river will rise,” says Mr. Asselin. This state-of-the-art tool will help emergency measures workers be “more responsive,” he says. “Currently, we rely [à leurs] knowledge, which is good, but which does not allow us to simulate in reality where the water will go if it ever overflows. »
A new approach
This real-time, large-scale simulation project is possible thanks to the data provided by the City, which makes it possible to accurately reproduce the buildings and their location.
But the goal of Professor Poullis and his team is to be able to use simulation in as many places as possible, and above all, on a large scale, which traditional three-dimensional modeling methods do not allow.
“Since the 1970s, people have been trying to reconstruct 3D objects from images. There are very robust techniques,” which work very well on a small scale, explains the researcher, “but as soon as you try to talk about 200 megapixel images, these techniques fail miserably”.
Traditional techniques locate key points in images and seek to associate them with other distinct points. But when it comes to reproducing an entire city, “there are not as many distinctive points as there are areas to reconstruct”, and too many elements may look alike, which blurs the reconstructions.
More modern deep learning techniques are also not suitable. They, too, work “very well for small images”, but “the biggest downside is that you have to train your model”. So, as soon as it is applied to an area where “the architecture is different, you have to [tout] restart “.
In order to create “digital twins” on a large scale, Mr. Poullis and his team have developed a new modeling technique, called HybridFlow. In addition to relying on distinctive points between images, as traditional methods do, HybridFlow identifies areas grouping together several pixels, which coincide on the different shots.
This technique notably made it possible to model the city of Montreal, thanks to images taken by an airplane flying at an altitude of more than 9,000 meters.
Multiple Perspectives
The simulation of natural disasters is the most interesting use of this technology, says Poullis. But it can materialize in many other applications.
“Uses range from underwater archeology in virtual reality, to [développement de simulations pour] improve the social skills of people with autism,” rejoices the director of the immersive and creative technologies laboratory at Concordia University.
It was indeed possible to virtually reproduce a particularly dangerous pedestrian crossing, to “help autistic children learn to cross the street. Obviously, trying to do this in a [vraie] street is very dangerous. And we have seen a significant improvement in their performance,” says the professor.
This content is produced in collaboration with Concordia University.