We detected and classified roadside infrastructure into five categories, building a comprehensive, geolocated digital inventory of the assets that line the network.
The challenge
The client lacked an up-to-date, spatially accurate map of roadside assets. Field surveys were costly, slow and often incomplete, with no automated way to tell, say, a curbstone from a guardrail.
Our approach
Deep-learning models over high-resolution aerial or mobile-mapping imagery detected, classified and geolocated five feature classes:
- Curbstone — raised edges separating roadway from sidewalk or gutter.
- Fence — roadside barriers in metal, wood or mesh.
- Guardrail — impact-absorbing barriers on the roadside or median.
- Paved / unpaved border — the transition where asphalt or concrete meets gravel, soil or vegetation.
- Pole — utility, streetlight, sign or camera masts.
Results & benefits
- A complete digital inventory of roadside assets across the network.
- The furniture-and-boundary layer that rounds out an HD map for safety and asset management.
