"Will it read my plates?" is the right question, and it almost never gets a straight answer, because the honest one is a question back: where were you planning to put the camera?
Licence plate recognition disappoints more often than most security technology, and when it does the postmortem usually finds a capable camera mounted somewhere it could never have succeeded. The system was not oversold on capability. It was undersold on geometry.
The useful thing about that is you can check the geometry before you spend anything.
Plate reading is not a camera problem
The instinct is that reading plates is hard, so it needs a better camera, so the fix for poor results is a more expensive one. That instinct produces a lot of wasted money.
What the software needs is a plate that is big enough in the frame, square enough to the lens, and captured at a moment when it is legible. Those are properties of the mounting position, the lens choice, and where the vehicle is when the picture is taken. A camera that can deliver all three from a good position will outperform a far better camera in a bad one, and no amount of resolution rescues a plate seen at a sharp angle.
Manufacturers publish what those requirements are. They are not secret, they are not vague, and you can hold your site against them today.
The published numbers
The figures below are Axis's, for their plate reading analytic. Other vendors publish their own and they are broadly comparable, but treat these as one manufacturer's stated requirements rather than as a universal standard — if you are buying a different platform, ask for that platform's numbers and check them the same way.
Resolution on the plate. Axis specifies that "the width of the license plate equals at least 130 pixels for license plates with one row and at least 70 pixels for license plates with two rows."
BC passenger plates are single-row, so the number that applies here is 130 pixels across the plate. Note what this is measuring: not the resolution of the camera, but how many of its pixels land on the plate itself. That is the concept covered in more depth in our article on pixel density, and it is the thing a specification sheet's megapixel number cannot tell you on its own.
Angle. "The camera's mounting angle should not be larger than 30° in any direction." This is the requirement that most installations break, and it breaks in two directions at once — cameras get mounted high (steep vertical angle looking down at the plate) and off to one side (horizontal angle across the lane). Each may look acceptable alone. Together they compound.
Tilt. "The image of the license plate should not tilt more than 5° horizontally." This is about rotation in the frame rather than viewing angle, and it is usually a mounting-bracket problem: a camera set slightly off level renders every plate on a slant. It is also the cheapest of all these problems to fix, and the one most likely to survive for years because nobody looks.
Distance. Axis gives two bands depending on what the system is for. For an access-control scenario — a gate or a barrier, where the vehicle stops or nearly stops — the capture distance is 2–7 m. For free flow, where traffic keeps moving, it is 7–20 m.
Height, and the rule that ties it together. For access control the camera goes up to about 3 m; for free flow, roughly 3–10 m. And the relationship that makes all of it coherent: capture distance should be approximately twice the mounting height.
That last one is the single most useful sentence in this article. It converts an abstract angle requirement into something you can check with a tape measure. Mount at 3 m, capture at about 6 m. Mount at 5 m, capture at about 10 m. If your intended camera position is 6 m up and the vehicles pass directly beneath it, the geometry has already failed and the camera model is irrelevant.
Why the two scenarios are genuinely different
The split between access control and free flow is not marketing. It changes what is achievable.
At a gate or barrier, the vehicle is slow or stopped, its position is constrained to within a metre or so by the lane, and you know where the plate will be. This is the easy case and the accuracy people expect from LPR is realistic here. It is also the case with the highest value, because the read is usually driving a decision — open the gate, log the arrival, flag a vehicle that should not be there.
In free flow, the vehicle is moving, it can be anywhere across the lane, and it may be beside another vehicle that hides it. Every requirement above gets harder to hold simultaneously, and the achievable read rate drops. It can still be worth doing. It should not be quoted to you at gate-level expectations.
If someone proposes a single camera to cover both a gated entrance and a general parking area, that is one camera being asked to satisfy two different geometries. It will do one of them.
Conditions that degrade a read
Axis names the one that surprises people: "direct sunlight can distort the image, for example, during sunrise and sunset."
This is worth dwelling on because of how it presents. A system commissioned in the afternoon in October can develop a daily failure window months later, when the sun's angle changes and lines up with a camera facing the wrong way for twenty minutes each morning. Nobody connects the two events. The system is simply described as unreliable.
Ask which way each camera faces, and ask what happens at low sun. On an east-facing entrance in BC, this is not a hypothetical.
Plate condition is the other honest limit, and it is entirely outside the system's control. A plate obscured by road spray in February, held in a dealer frame that crops its edges, dented, faded, or mounted behind smoked plastic, is a plate the software may not read. No installation fixes that, and any vendor promising a read rate without qualifying it against real winter plate condition is quoting you a laboratory number.
How to check your site in fifteen minutes
Before you take any quote seriously, do this yourself.
Stand where the camera is proposed to go, and look at where a plate will be. If you are looking steeply down, or across the lane at a sharp angle, stop there — that is the 30° requirement failing, and it is visible without instruments.
Measure the height and the distance. Divide the distance by the height. If the answer is close to 2, the geometry is in the right region. If it is well under 2, the camera is too high for where the vehicles are. This one check catches the majority of doomed installations.
Look along the sight line at eye level for the camera. Is there a bollard, a sign, a hedge that grows, a gate arm that sits in the frame when raised, a parked vehicle that habitually blocks the view? LPR needs an unobstructed look at a small object; a partial view is a failed read, not a degraded one.
Go and look at sunrise, and at sunset. Or at least establish which way the camera faces and be honest about what the sun does there.
Ask what happens on a failed read. This is the design question rather than the site question, and it separates a system that works from a system that is merely accurate. Does a missed plate mean a driver sits at a barrier with no way in? Is there an intercom, a call button, a manual override, an attendant? A gate whose only entry mechanism is a plate read has made an accuracy figure into a single point of failure, and even excellent LPR will fail sometimes.
The honest expectation to hold
Well-sited plate recognition at a controlled entrance, on plates in reasonable condition, is a mature technology that does what it says. Poorly sited plate recognition is unreliable in a way that no support call ever fully fixes, because the problem is structural and the fix is a lift, a new mounting position, and possibly a different lens.
The difference between those two outcomes is decided before anything is purchased, in a conversation about where the camera goes. If a quote for LPR arrives without a mounting height, a capture distance, and a stated angle, it is not yet a design — and the question to send back is simply: at what height, at what distance, at what angle, and what happens when the sun is low?
