Tracking 13,000+ public road projects with geotagged photos and machine learning
Philippine government agencies each kept their own photos and data on the roads they funded. Symph built one open platform to track every project, with a model that classifies road surfaces across 200,000+ images.
Road projects tracked
Road images classified
Road classification accuracy
We are a software development team trusted for 14 years by established companies and startups.
About the client
The Department of Budget and Management (DBM) prepares the Philippine national budget and oversees how public funds are spent. As spending on local roads grew, so did the need to show where the money went and what it built.
The challenge
Four government agencies were collecting geotagged photos and data on road projects. Each kept its own records, and the agencies rarely shared them.
Classifying a road’s type and condition meant a person looking at photos one at a time, and there were hundreds of thousands of them.
The solution
Symph ran workshops with the agencies and the World Bank to map how road data was collected and checked. Then it built Geostore, an open-source platform with four parts.
- A dashboard where agencies upload project details, photos, and other data, and track each project.
- A public viewer, where anyone can browse the projects.
- Geoprocessing that turns each upload into route, surface type, surface quality, and road width, as tables and as map data.
- A machine learning classifier, trained on about 10,000 photos labelled by people, that classifies road type and quality across the rest.
Symph also ran the product study and strategy design that shaped the build.
Impact
13,000+ road projects in one place
Road data from four agencies sits in one shared repository instead of four separate ones.
200,000+ images classified at 84% accuracy
The model took over a job that people had done photo by photo.
Open to the public
Anyone can browse the projects in the public viewer and check what public road money built.
