Machine Learning for Improved Services: Potholes
Memphis embraces opportunities to revolutionize service deliveries.
Memphis has 6,800 lane-miles of city streets, enough to drive back and forth to Los Angeles four times. With all those miles come the potential for potholes – a lot of potholes. Last year, Public Works crews repaired some 63,000; roughly 7,500 of them were ones reported by citizens to 311. The rest are potholes the crews encountered on their own. With this process, it is easy for potholes to go undocumented until they become so egregious they are then reported. We began to ask, “Can Public Works benefit from artificial intelligence in the fight against potholes?”
In February 2019, the City of Memphis began a partnership with Google and SpringML to pilot machine learning (ML) to improve city services. Pothole detection using TensorFlow technology is a primary focus. TensorFlow supports ML and artificial intelligence (AI) by providing an open source software library for high performance numerical computation, thus making it easier to build and deploy ML powered applications.
A resource library was compiled and included select video from MATA buses according to routes having the most potholes reported, identified through 311 data. Stand up meetings were held three days a week over a five-week period. These meetings were used to identify, collect, and share data to refine the ML model.
As of late April, the ML model has an accuracy of 84%. As more data is collected and the model refined, this number will grow.
The next step is to standardize the data capture. Google and SpringML will work with the City to create a standard for camera mount and angle, and the ML model will be updated to reflect customized data classifications tailored to the City’s needs.
I’m thrilled about this opportunity for Memphis through our partnership with Google and SpringML. This is the tip of the AI iceberg when it comes to improving services through data. I’m excited to see all the ways Memphis can capitalize on this technology and I look forward to the journey ahead!
Contact the Indianapolis Mayor and opposing candidates. They would likely jump on anything tangible coming from this work. At least Indy would make for a perfect bulk test of your technology!
Please come to San Diego, it rained for a few days during the "winter" and now the streets fall apart. They also fall apart without rain.
I’ve been glad to see how quickly potholes are filled in after reporting them on 311. Thanks to the folks getting the job done. I’m glad to see the use of tools like ML to make a difference, too.