An AI-Assisted Mapping Tool for Disaster Management with Humanitarian OpenStreetMap

Objectives

• To develop an AI-assisted mapping tool to support humanitarian organisations in mapping disaster-affected areas.
• To improve the accuracy and efficiency of mapping efforts, especially in remote or inaccessible areas.

Methodologies

• Convolutional neural networks (CNNs)
• Transfer learning
• RAMP deep neural network
• HOTLib Python library

Outcomes

• A deep neural network model that can accurately predict building footprints based on aerial images with an average accuracy of around 94%.
• A Python library (HOTLib) that provides a reference implementation of the AI-assisted mapping workflow.

Impact of the solution

• Improved accuracy and efficiency of mapping efforts in disaster-affected areas.
• Increased availability of map data for humanitarian organisations, which can lead to better decision-making and more effective relief efforts.

Conclusion

This project developed an AI-assisted mapping tool that can help humanitarian organizations improve the accuracy and efficiency of their mapping efforts in disaster-affected areas.

Find a detailed technical case study on the project here.

Final presentation