An AI Assisted Collaborative Mapping Tool for Disaster Management

Challenge Completed!

The team developed a deep neural network for the Humanitarian OpenStreetMap Team (HOT) and their open-source AI-assisted mapping tool where AI models are created and trained by end users in a live environment. 

The problem

Each year, disasters around the world kill nearly 100,000 and affect or displace 200 million people. Many of the places where these disasters occur are literally ‘missing’ from any map and first responders lack the information to make valuable decisions regarding relief efforts. 

There is a need for map data that revolutionizes disaster management, reduces risks, contributes to the achievement of the Sustainable Development Goals, and helps humanitarian organizations that are trying to meet the needs of vulnerable people.

Our Partner – HOT develops open-source apps and tools for collaborative mapping and geospatial data collection. Their tools are free for all to use and leveraged by partners such as Red Cross societies, Médecins Sans Frontières, UN agencies and programs, government agencies, and local NGOs and communities.

The project outcomes

The Humanitarian OpenStreetMap Team (HOT) utilizes OpenStreetMap (OSM) for its humanitarian actions and community development. The spatial location of buildings is essential information in terms of disaster resilience. Therefore, the task in this challenge was the development of an AI-assisted mapping tool supporting the HOT mappers in the delineation of building footprints based on aerial images.

The project team tested several architectures of Convolutional Neural Networks (CNN) during the challenge. As the provided input data was limited, the final approach was to apply the pre-trained model from the RAMP project. This deep neural network was trained with large amounts of satellite images and was used for transfer learning. The RAMP model achieved the best prediction results during fine-tuning with an average accuracy of around 94 % and an Intersection over Union score (IoU) of around 84 % across all test regions with an average inference time of ~1.7 seconds.

One major outcome of the challenge was the development of the HOTLib library as a reference implementation of the planned workflow. This Python library supports HOT in the preprocessing of the input data, the inference step based on the RAMP model as well as the post-processing of the prediction results. These project results will enable HOT to provide their mappers with an AI-assisted collaborative mapping tool.

Find a detailed technical case study on the project here.

Final presentation