Accuracy is decided at the roadside
When a corridor deployment misses plates, the first instinct is usually to look at the recogniser. It is almost always the wrong place to look. Recognition on a modern LPR camera is good enough that the limiting factor is the frame it was given, and the frame is a function of where the camera was bolted and what it was pointed at.
This matters commercially as well as technically. Recognition runs at the camera in a Vaultris deployment, which means the model of camera and its placement decide read quality far more than the central hardware does. You cannot buy your way out of a bad mounting position with more servers.
Angle is the first constraint
A licence plate is a retroreflective rectangle. It returns light efficiently when the camera is close to square-on, and poorly as the angle opens up. Two angles matter and they compound.
- Horizontal angle — how far off the vehicle's line of travel the camera sits. Keep it under about 30°. Past that, characters begin to compress toward the edge of the plate and the recogniser starts producing confident wrong answers rather than obvious failures.
- Vertical angle — a function of mounting height and how far up the road the capture zone sits. The same 30° guidance applies. Mounting high to get a clean view over traffic and then capturing close to the pole is the most common way to blow this budget.
The capture zone is sized by speed, not by lens
The recogniser needs a minimum number of usable frames of the plate. That gives a capture zone — the stretch of road where the plate is large enough, sharp enough and square enough to read.
The faster the traffic, the less time a vehicle spends in that zone, and the more the shutter has to be shortened to stop motion blur. A shorter shutter needs more light, which is why fast corridors need stronger illumination rather than a wider lens. Zooming in to make the plate bigger shrinks the capture zone at the same time, which is a trade that catches people out.
Free-flow highway lanes, a slow toll plaza approach, and a signalised junction are three different problems. A single specification applied across all of them will underperform in at least two.
How many lanes one camera can carry
At low speed and shallow angle, one camera can cover more than one lane. At highway speed it generally cannot, for a reason that is geometric rather than budgetary: covering the far lane opens the horizontal angle, and the far lane is also where a vehicle in the near lane is most likely to occlude the plate entirely.
Plan one camera per lane where free-flow reads matter, and treat multi-lane coverage as an explicit compromise with a stated expected read rate rather than as the default. It is far cheaper to add a camera at build time than to explain a coverage gap after an incident.
Light is a design input, not an afterthought
LPR cameras generally use their own infrared illumination and a filter that rejects most visible light. That is deliberate: it makes the read independent of street lighting, headlights and time of day, and it is why a well-specified plate camera produces a stark monochrome image rather than a pretty colour one.
Two consequences follow. First, the illuminator has range limits, so the capture zone has to sit inside them. Second, a plate camera is not an overview camera — it will not tell you the colour of the vehicle or what the driver was doing. Deployments that need both attributes and plates need both cameras, and the sizing has to say so.
What to check before you pour concrete
- Sight line at the intended capture zone, at the intended time of day, with traffic present rather than on an empty Sunday morning.
- Occlusion from signage, gantry legs, vegetation that will be larger in three years, and high-sided vehicles in the adjacent lane.
- Vibration. A gantry that moves under load will blur a short-shutter frame. Mounting mass and position matter more than they look on a drawing.
- Power and backhaul at the position you actually want, not the position where the duct already runs.
- Maintenance access — a camera that needs a lane closure to clean is a camera that will stay dirty.
Placement changes the central build too
Read volume per camera is the input to every downstream figure: events per second, database growth, image storage, and the retention bill that follows. A corridor with cameras on every lane produces materially more reads than one with shared coverage, and the sizing has to be built from the placement plan rather than from a camera count.
Our hardware and sizing page works through that arithmetic tier by tier, including what each retention period costs once the reads start arriving.