Road/Extraction

Blog/ 2026-10-26

What a turn restriction is, and why realignment breaks it

A winding two-lane road curving past a hillside with a road sign and guardrail at the bend.

A turn restriction is an attribute on an intersection, not on a road. It says something like "no left turn from this approach onto that one" or "right turn only," and it's tied to a specific node in the routing graph along with the edges that meet there. That sounds simple until the intersection itself moves.

Where the restriction actually lives

In most graph schemas, a turn restriction is stored as a relation between an incoming edge, a node, and an outgoing edge. Some engines store it as a maneuver on the node directly, others as a restriction object referencing way IDs on both sides. Either way, the restriction only knows the IDs: edge A into node X, then out edge B, forbidden. Change any of those three IDs and the restriction either fires on the wrong pair of edges or stops firing at all.

The part that trips people up: a turn restriction is attached to whatever edge and node IDs your graph assigned the last time someone digitized that intersection. It has no tie to "the intersection of Route 9 and Mill Street" as a place. Rename the road, and the restriction survives fine, because IDs don't care about names. Move the intersection fifteen meters north because the county realigned the approach, and the restriction can break even though nobody touched the attribute table.

What realignment actually does to the graph

A road realignment usually means a new centerline, a shifted node, or both. The approach angle changes, a slip lane gets added or removed, a signal gets relocated, a median gets extended to kill a movement that used to be legal. None of that shows up as an edited attribute. It shows up as new geometry that your extraction or digitizing pass has to turn into new edges and a new node, and the old turn restriction relation was written against the old ones.

Three things commonly happen after a realignment lands in your source imagery but the restriction logic doesn't get rebuilt:

A restriction references an edge ID that no longer exists on the ground, so the routing engine silently drops it. Drivers get routed through a turn that's now illegal, because the realignment added a median or a no-left sign that your graph never heard about.

A restriction survives on an edge ID that's technically still present but now represents a different physical movement, because the realignment changed which approach feeds which exit. The restriction is "live" in the data and wrong in the field.

A new slip lane or channelized turn gets added as a new edge with no restriction relations at all, so a movement that should be forbidden (say, a right-on-red ban at a redesigned intersection) just isn't encoded anywhere.

None of these show up in a standard geometry diff that only checks whether a line segment moved. You need something that flags the intersection as changed and pushes production to rebuild the maneuver logic there, not just redraw the centerline.

Why this matters more than a missing road segment

A missing road segment usually produces an obviously bad route: a detour that's too long, a dead end. A stale turn restriction produces a route that looks perfectly reasonable and sends a driver into a movement that's been illegal since the DOT repaved the corner. That's the complaint that lands on a support queue with no obvious cause, because the geometry looks fine on a quick visual check. Someone has to notice the realignment, trace it back to the turn relation, and manually correct it, often long after the construction crew finished.

This is also why turn restriction upkeep tends to lag behind centerline upkeep in most production shops. Redrawing a segment is a geometry task. Rebuilding restrictions is a logic task that requires knowing an intersection changed in the first place, which is exactly the kind of change that's easy to miss between survey cycles. Road Extraction is built around catching that gap: it flags new and changed segments from current high-res imagery on a quarterly cadence, so a realigned intersection surfaces as a change to merge and re-check rather than something your team finds out about from a routing complaint.

If your routing graph has been running quiet on turn logic for a while, it might be worth checking which of your recent realignments actually got their restrictions rebuilt.

Start a project