PRACTICAL AI
Artificial intelligence could help city maintain roads, save money


CBC
Mon, October 23, 2023 


Initial tests by the City of Calgary in using artificial intelligence (AI) show it can do a better job of detecting potential potholes, cracks and other road defects than humans can.

Earlier this year, officials developed a system that uses AI machine-learning technology to gauge pavement conditions.

It was a joint project between the city's mobility department, its information technology department and AltaML, an Alberta-based company that designs and implements AI solutions for businesses.

An IT engineer with the city, Ashiq Rahman, said they first had to tell a computer algorithm what kinds of defects it should look for in the imagery.

"You throw this training set into the model and the model, from that set, will learn these defects and when it learns, it means the model will become ready," he said.

"Then you can use this model to test any new defects that your camera will capture."

What the AI program was able to do had everyone on the project feeling pretty excited.

To illustrate how it performed, the city has produced some images. In blue boxes on the pictures, you can see potholes or cracks that are detectable by the human eye.

But there are also red boxes which outline the deformations detected by AI with varying degrees of accuracy.

"It means the algorithm is saying here that 'look i see a defect here and I am 98 per cent confident that this is a road defect that you are trying to detect,'" said Rahman.

It means city crews could repair a defect before it becomes a bigger problem or they could monitor the situation and deal with it as it becomes visible.

The director of the city's mobility department, Troy McLeod, said having this kind of intelligence can generate all kinds of dividends.

"It can predict where those defects will propagate and come up with a priority repair program so that we can address deficiencies in advance of any failure or any further deficiencies in the roadway," said McLeod.

"This will help us save costs on repair for our pavement assets."

So far this year, the city has patched about 25,000 potholes. That program costs more than $6 million annually. The city also typically spends about $40 million a year repaving roadways where the pavement has reached the end of its lifecycle.


The kind of video that's been gathered using a new high-resolution 11K camera that can be fed into an AI program is shown. (Mike Symington/CBC)

McLeod said the next big step in their AI work will come in January 2024. That's when the city plans to start mounting special high-quality cameras on some of its vehicles.

The devices will be used to gather high resolution, 360-degree video as the vehicle is driven along city streets.

That higher quality video will be fed into the AI program to further test its effectiveness at spotting current and potential problems in the asphalt.

An imagine showing potholes visible to the human eye and ones that were detected by an AI program. (The City of Calgary)

The chief information technology officer for the City of Calgary, Jan Bradley, said the work is "incredibly exciting."

She said that while the discoveries so far have been good news for road maintenance, there is a lot more potential for the city.

"The opportunities that we have to leverage artificial intelligence, machine learning, to really solve business unit use cases that can be very time intensive, labour intensive, if we can transition that work to some of these AI models, I think the opportunities are endless."

For example, AI could point out city trees that need watering based on their deteriorating condition, or pinpoint illegal temporary signs next to roads.

Bradley said the city has been working in this field for a few years, but a more recent development has boosted their progress.

"The launch of ChatGPT and the public awareness has really allowed us to accelerate how we are going to be able to use and leverage this technology in our organization. And that's what we're really excited about."

For those who worry about a growing number of city vehicles on the street with cameras, Bradley said there are rules to protect the public's privacy.

She said the city won't be intentionally collecting or keeping imagery of people's faces or licence plates. The focus will be gathering video for what's needed for the AI model to do its work.

The city believes it is at the forefront of this research among Canadian municipalities.

McLeod said he's aware of other cities that are doing AI work regarding potholes, but not related to cracks and other road defects.