Output of network
Input into trained network
We are leveraging the ubiquity of smartphones and availability of municipal vehicles to automate and objectively assess city roads.
Check us out at RoadReader.io
Potholes. We all hate them.
Every year in the U.S., damage to vehicles from potholes costs Americans $3 billion to fix, and road-related conditions account for 52% of traffic fatalities. Today, 55% of San Diego's roads are ranked as poor. Currently, our city has limited means of detecting and reporting road conditions objectively and at scale.
We think that we can modernize the way roads are assessed with just a smartphone! What we came up with is RoadReader, a platform that leverages image detection technology and sensory data to automate the classification and reporting of road hazards.
How We Do It
It works by mounting smartphones with our app open on the dashboard of municipal (and crowd-sourced) vehicles. Road data is then passively collected and processed through our neural network which detects hazards like cracks, potholes, and line blurs, as well as road friction. Within 24 hours, these road conditions will appear on our self-updating map for city officials to then take action on.
We believe that San Diego having access to a self-updating assessment of its roads could be invaluable to the city and its residents. With this technology, city officials and dispatchers will be able to react quicker and more effectively to road hazards as they appear, resulting in smoother, safer roads for all. Additionally, San Diego will gain the ability to be proactive in its road maintenance; rather than spending money and resources fixing a fully-formed hazard, they can locate ones that are just beginning to form and deal with them, decreasing the cost-to-fix by upwards of 700%.
What We Have So Far
We have a neural network trained to detect cracks, potholes, and line blurs present in video or image. We also have an app to collect relevant road data through video, accelerometer, and gyroscope sensors, as well as a database to store the data and process it through our network.
We are working to create a map UI that displays road hazards detected with our neural network, and to increase the accuracy and types of hazards our model can detect. We hope to partner with the city to launch our project and mount phones onto a number of their vehicles. We would also like to crowd-source our data collection to the public through incentives, such as friendly competitions between universities, or a raffle reward system for those who collect road data through our app.
Please check us out at