Open AI Caribbean Challenge: Mapping Disaster Risk from Aerial Imagery

Can you predict the roof material of buildings from drone imagery? Leverage aerial imagery in St. Lucia, Guatemala, and Colombia to more accurately map disaster risk at scale. #disasters

$10,000 in prizes
dec 2019
1,419 joined

Castries drone imagery

The combination of machine learning with high resolution drone data will be instrumental in community resilience. We look forward to the Flying Labs network leveraging the winning algorithm to increase the resilience of their local communities.

— Joseph Muhlhausen, Head of Drone & Data Systems at WeRobotics

Why

Natural hazards like earthquakes, hurricanes, and floods can have a devastating impact on the people and communities they affect. While buildings can be retrofit to better prepare them for disaster, the traditional method for identifying high-risk buildings involves going door to door by foot, taking many weeks if not months and costing millions of dollars.

The Solution

Roof material is one of the main risk factors for earthquakes and hurricanes, and a predictor of other factors like building material that are as not readily seen from the air. The World Bank Global Program for Resilient Housing and WeRobotics teamed up to prepare aerial drone imagery of buildings across the Caribbean annotated with characteristics that matter to building inspectors.

Then in this competition, participants worked with this data to build their best rooftop classifiers. Computer vision models that most accurately map disaster risk from drone imagery will help drive faster, cheaper prioritization of building inspections and target resources for disaster preparation where they will have the most impact.

The Results

The top algorithm beat out more than 2,700 other submissions. From among 5 categories of construction materials, this solution correctly identified roof types 87% of the time, more than twice the pre-competition benchmark! This and all the other prize-wining solutions were shared openly for continued learning and development. Check them out below.


RESULTS ANNOUNCEMENT + MEET THE WINNERS

WINNING MODELS ON GITHUB