Bad infrastructure causes trillions of dollars in lost productivity. We thought about why it was so common to see deteriorating infrastructure, like potholes in roads and how none of us have ever really done anything to help fix the problem. We did some research and found out that the only way to report damaged infrastructure was to call a special number and it was clear that we could find a better way to help infrastructure planners make better, more data-driven decisions.
What it does
Our project is a crowdsourced data platform for damaged infrastructure. A simple to use iOS app allows anyone to automatically report damaged infrastructure one encounters on the road. Using a deep convolutional neural network, our app automatically detects damaged roads and uploads a report to a database that includes the image of the damaged infrastructure and the geolocation of the damage. The app also allows the user to use a voice prompt to manually submit a report. We have also built a web application that allows infrastructure planners to visualize and gain insights from the data platform. The app includes a heat map of where damaged infrastructure has been reported and a simple table that includes data from reported damage.
How we built it
The iOS app takes images during a user's drive and queries a deep neural network served on Google Cloud Machine Learning Engine in order to determine if damaged infrastructure is present in the image. If so, the app uploads a report containing the image and the location of the damage to a Firebase database. The web app interfaces with the database in order to present a heat map of infrastructure damage through the Google Maps API and a table of all the relevant data of the reports.
Challenges we ran into
Getting our model in the right format to deploy on Google Cloud Servers was a challenge. Getting the iOS app, web app, and database to seamlessly work together was also a challenge.
Accomplishments that we're proud of
We are proud that we were able to create a platform that can give anyone more agency in the infrastructure planning process and more relevant insights to infrastructure planners. Our project can hopefully help reduce the inefficiencies caused by poor infrastructure, and therefore reduce the negative impact humans have on the environment.
What we learned
We learned that technology can enable anybody to contribute to big problems and that even small collective efforts can potentially make a big impact. On the technically side of things, we learned a lot about deploying applications on the web and interfacing many technologies with each other.
What's next for DeepHole
We hope that our project can help improve infrastructure planning and repair. In the future, we would like to see the deep learning model used to be improved even more by being able to detect more types of infrastructure damage and more accurately. With the rise of autonomous cars driving more always-on cameras into cars, we hope that a system like the one we built could be directly integrated into all cars, allowing for seamless reporting of infrastructure in need of repair.