If you’re in government, it’s time to start considering the value of machine learning. One of the more persistent issues all Mayor’s face is how quickly potholes can be detected and resolved. Memphis is no different – what is different is how we decided to address the problem.
City of Memphis
Memphis is the logistics hub of America and has 6,800 lane-miles of city streets. FedEx and more than 400 trucking companies operate out of the city because of its central location, less than 500 miles from the mean center of US population.
In Memphis, pothole reporting is a manual process. Residents call in the more severe potholes and Public Works crews report the most, about 90%. The existing reporting process does not provide a complete list of potholes in the city, nor does it predict where new ones will form. That’s why the City of Memphis partnered with Google and Spring ML to pilot machine learning to locate and classify potholes.
Using Machine Learning to Locate and Classify Potholes
Spring ML used low-resolution video from MATA buses for the proof of concept. Working in a short timeframe and using low-resolution video, they delivered a pothole detection accuracy of 84%. In addition to the proof of concept, Spring ML developed a user interface that connects to our current pothole database. This UI allows potholes to be effortlessly entered into the existing 311 system.
“Working in a short timeframe and using low-resolution video, they delivered a pothole detection accuracy of 84%.”
Making Memphis Government More Efficient
Because of this pilot, the city is working to standardize cameras, train city employees to refine the model, and increase focus on ML for improved resident experiences. The adoption of this technology makes Memphis government more efficient, alleviates the burden of citizen reporting, and positions us to be proactive in our battle against potholes.
