How to Inventory Road Markings from Aerial Imagery

Most agencies still build their marking inventory one of two ways: a crew drives the network with a van-mounted retroreflectometer and logs condition point by point, or someone on the GIS team opens the latest ortho tiles and starts tracing centrelines by hand. Both work. Both also eat weeks you'd rather spend on the rest of the asset register, and the digitizing approach drifts in quality depending on who's holding the mouse that week.

There's a third option that's been maturing alongside satellite and aerial resolution: extract the marking layer directly from overhead imagery using feature detection instead of a human tracing every line.

What gets extracted, and what it looks like in the register

From VHR imagery at 0.3 m GSD or finer, a marking extraction pass can pick out the line work that actually matters for an asset register: centrelines, edge lines, lane lines, stop bars, crosswalks, turn arrows, and parking bay striping. Carriageway and footway extents come out of the same pass, since markings don't mean much as a layer without the pavement they sit on.

The output you want isn't an annotated image. It's GIS features: polylines for the linear marking types, polygons for crosswalks and bays, each one carrying geometry you can drop into your existing schema. If your register already has fields for marking type, feature class, and segment ID, the extracted layer should snap into that structure rather than forcing you to invent a parallel one. That's the difference between a one-off mapping exercise and something your asset management system can actually ingest.

Centreline extraction specifically

Centreline marking extraction gets its own mention because it's usually the layer every downstream process references, lane configuration, striping renewal schedules, pavement marking retroreflectivity programs tied to MUTCD or state DOT standards. Pulling it from imagery means the geometry follows the painted line as it actually exists on the road, including the offsets and jogs that a schematic network dataset usually smooths over. When a resurfacing crew repaints a section slightly differently than the as-built drawing, the next imagery pass catches that, rather than the register quietly going stale until someone notices on the ground.

Where this fits next to a walking survey

Imagery-based extraction isn't a replacement for a detailed condition assessment. It won't tell you retroreflectivity values in millicandelas, and it's not going to flag a marking that's worn to a faint shadow under certain light. What it's good for is establishing and refreshing the geometric inventory, where is every striped feature, what type is it, how long is it, so your register has current, mappable features between the full field surveys that handle condition scoring. Run the imagery pass on a biennial cadence and you've got a standing layer that doesn't depend on whichever year the walking crew got funded.

For agencies working through a VHR satellite or aerial dataset in RGB, this is a workflow worth building deliberately rather than assigning to whoever's available. A workflow built specifically around pulling that inventory straight into GIS format is the shortest path from raw imagery to features your register can use without a re-digitizing pass in between.

If your register needs a marking layer refreshed without putting a crew back on the road, that's the gap Road Asset Inventory is built to close.