Pothole detection is a costly problem for VINCI Eurvia. Only a small part of the public takes the time to report potholes, and when potholes are reported, the reports are untrustworthy. This means that VINCI Eurivia inspectors have to do a large amount of work to regulary check for street potholes. We believe that an automatic system for live pothole detection would make the streets much safer and reduce the management costs for VINCI Eurvia.

The idea

Our main insight is to leverage the accelerometer data from thousands of daily commuters to predict the existence and the size of potholes. The results would be updated live in map in the management interface for the inspectors. The main challenge here is how to incentive the public to send us their accelerometer data.

For that we have created an app to score the driving quality of the users using accelerometer data. A good driver predicts early when he has to brake or accelerate. His driving style is comfortable and do not use much energy. In parallel, our app crowdsource the accelerometer data to predict potholes, and sends the events to our servers.

The prediction are clustered in our backend and shown live in map in a management web app for the inspector. The inspector can then check the suggested potholes and use the management app to remove or add them

How we built it

We developed the commuter app using react-native: We filter accelerometer data using a high pass filter to remove the effect of gravity. We use the amplitude of acceleration when accelerating in the different directions (acceleration and braking) to compute a score for the driving quality. We use a threshold over the norm of the acceleration to predict potholes. When a pothole is detected the gps coordinate and the amplitude of the acceleration is sent to the firebase database.

The VINCI inspector web app is developed in html+javascript using the HERE-API. The app reads the firebase database and shows the predicted potholes with number of occurrences. The inspector web app also allows to check and edit the data about predicted potholes.

Challenges we ran into

Processing and normalizing the accelerometer data to remove gravity effects and noise was challenging.

Accomplishments that we're proud of

Creating a fully working app for users and inspector using different technologies in less than 48 hours.

What we learned

Do not reinvent the wheel. Reuse solutions made by others, scientific publications are great source for (peer-reviewed) solutions. Divide and conquer. Monitor Analyse Process Execute very very fast, you only have some hours.

What's next?

We believe that the main value of our app lies in the access of accelerometer data of thousands of users. This data has a high potential for a multitude of scenarios in addition to pothole detection. We could leverage the accelerometer data using machine learning for: live traffic analysis or analysis of driver behavior in different portion of the road (which could be used for determining advertisement placement, improving road and infrastructure design..)

But first, we would have to calibrate our driving scores and pothole detection algorithms using live data. We have already found public data for thousands of driving hours including video and accelerometer data.

Make a better icon :)

Built With

Share this project: