The challenge
African road infrastructure suffers from accelerated degradation, compounded by the lack of preventive monitoring systems. Manual detection of potholes and road damage is costly, slow and unscalable. The challenge: design a system capable of automatically detecting road degradation from satellite imagery at continental scale.
Project origin
GeoSmart Vision is the competitive materialization of TerraPulse Vision, a research topic I formulated in 2023, motivated by a road accident I suffered in 2020 caused by a pothole hidden under mud. In 2025, my university's FabLab was looking for a topic on mapping and territorial monitoring to participate in MASS 2025. I proposed TerraPulse Vision, convinced of the relevance of the problem it solved: reducing accident risks and saving lives. The university validated the proposal, renaming it GeoSmart Vision. I remained the project lead and owner throughout the competition.
Methodological approach
As project lead, I led a multidisciplinary team following a structured methodology: full authorship of the technical specifications document, design of the image processing pipeline, and development of the machine learning model. The approach combines satellite imagery with ground truth data to train an automatic pothole detection model.
Technical architecture
The system is organized in three layers. Input: satellite imagery cross-referenced with OpenStreetMap data. Processing: a pipeline integrating a machine learning model for detection and geolocation of road degradation. Output: a web cartographic platform enabling real-time visualization of degraded zones and impact assessment.
The deliverable
A functional prototype demonstrating automatic road degradation detection, complete with full technical documentation and a live cartographic visualization interface.
My contribution
Original research proposal (TerraPulse Vision). System architecture design. Machine learning model development. Full authorship of technical specifications. Team leadership and deliverable coordination.
Recognition and impact
The project won 2nd Prize at the African Market for Space Solutions 2025, recognized for its technical robustness and innovation. GeoSmart Vision constitutes the scientific core of Road Intelligence, the GEOWATCH branch dedicated to road monitoring.
Technical resources
Technology badges are displayed below.
- Python
- TensorFlow
- scikit-learn
- Sentinel-2
- PostGIS
- PostgreSQL