ANPR Camera Types: Choosing by Shutter, Form Factor and Power
Shape is the least useful way to sort ANPR cameras. The four axes that actually decide whether a site reads plates, the shutter arithmetic behind each speed class, and a decision table that maps speed, distance and power onto a type.
Published 17 November 2026 · 18 min read · Selection
Buyers sort ANPR cameras by shape: box, bullet, dome, all-in-one. Shape is the least useful of the four properties that decide whether a site reads plates, because two housings that look identical from outside can differ in something that matters more than resolution, which is the way the sensor exposes a moving plate.
This is a selection guide, not a definition page. It sets out the four axes that actually separate one ANPR camera from another, the arithmetic behind each one, and a table that maps a site's speed, distance and power situation onto a type. Every number below comes from the SZGOING range: a 5 to 40 metre read window on the 5MP models and 30 metres on the 4MP solar unit, 150 km/h capture on the high-speed models and 50 km/h on the 4MP units, 65 degrees of horizontal tolerance, a 1 TOPS on-board processor, H.265 at 1080p25, and an uplink that runs from 256 kbps to 8 Mbps.
Four axes, and only one of them is in the model name
Four properties separate one ANPR camera from another in a way that changes site performance. Everything else, including housing colour, bracket style and whether the label says AI, is procurement detail.
- Shutter type. Whether the whole sensor exposes at one instant or line by line. This decides whether a plate at speed comes out sharp or sheared.
- Where recognition runs. Inside the camera, on a recorder, or on a machine in a cabinet. This decides what fails when the link drops.
- Power and backhaul. Mains PoE, 12 V DC, 4G, or solar with 4G. This decides whether the location can be served at all.
- Speed and distance class. A unit built for a car park lane and a unit built for a tolling gantry are not interchangeable, even when both quote 1080p.
The trap is that only the fourth axis tends to appear in the model name, and only the first is invisible in a product photograph. A buyer comparing two 5MP cameras on resolution, lens range and delivery time is comparing two of the properties that matter least, and skipping the one that decides whether the installation works at 100 km/h.
| Axis | What it changes on site | What to check |
|---|---|---|
| Shutter type | Whether characters survive motion; whether vertical strokes lean in the frame | Sensor specification, or a frozen frame of traffic at site speed |
| Where recognition runs | What still works when the network fails; how much data leaves the site | On-board processor, protocol list, local storage |
| Power and backhaul | Whether the location can be served at all | PoE or 12 V input, cellular module, solar panel rating |
| Speed and distance class | Lens range, exposure ceiling, trigger precision, lane count | Rated speed, read window, lens range, lane count |
The four properties that separate one ANPR camera from another
Shutter type: global versus rolling
A global shutter exposes every pixel for the same interval and ends that exposure on all of them at once. A rolling shutter exposes and reads the sensor line by line, from one edge to the other. On a stationary scene the two produce the same picture. On a plate moving at speed they do not, and this is the single most common reason a well-aimed installation underperforms its specification without ever showing an obvious fault.
Start with exposure time, because it sets the blur floor for either type. A vehicle at 150 km/h covers 41.7 metres per second, which is 41.7 millimetres per millisecond. At 1/500 s the plate travels 83 mm while the shutter is open. At 1/1000 s it travels 42 mm. At 1/2000 s it travels 21 mm. A Latin character on a standard plate is roughly 45 mm wide, so 1/1000 s already smears the plate by about one character width, and 1/2000 s brings the smear down to about half of one. That arithmetic is why high-speed lanes are specified at 1/1000 s or faster, and why the exposure ceiling of a camera is a more useful thing to check than its pixel count.
Now the part that is widely missed. Shortening the exposure does not remove rolling-shutter distortion, because the distortion does not come from the length of any single line's exposure. It comes from the fact that the top of the frame and the bottom of the frame are exposed at different moments. If the sensor needs tens of milliseconds to walk from its first line to its last, then the rows carrying the top edge of a plate and the rows carrying the bottom edge are sampled several milliseconds apart, and the vehicle has moved in between. Set that same rolling-shutter camera to 1/4000 s and each row is crisp, but the plate is still sheared, because the shear is a timing offset between rows rather than a blur inside them.
The magnitude is easy to underestimate. A few milliseconds of row-to-row offset at 41.7 metres per second is a displacement above a hundred millimetres, which is two character widths. On a frozen frame the effect reads as vertical strokes that lean in the direction of travel, characters that look taller at one end than the other, and a bounding box that sits slightly rotated against the plate border. Recognition confidence falls before recognition fails outright, which is the harder failure to catch: the camera keeps returning plates, a fraction of them wrong, and the log still looks healthy.
A ten-minute test that settles it. Park a vehicle with a clean plate at the working distance and confirm that the characters sit square in the frame. Then record traffic at the speed the site actually runs and freeze a frame. Pick a character with a strong vertical stroke, such as a 1 or an H, and compare its top and bottom against the plate border. If the stroke leans while the parked plate was square, the sensor is rolling and no exposure setting will fix it. If the strokes stay vertical and the image is simply soft, the problem is focus or light, and a settings change will help.
Where a rolling shutter is acceptable: lanes where vehicles stop or crawl, which covers a car park barrier, a weighbridge and a compound gate. The offset between rows scales with speed, so at walking pace the same camera that shears at 150 km/h produces a clean plate. Where it is not acceptable: free-flow lanes, tolling gantries and anything above roughly 60 km/h, which is the working threshold we use when specifying rather than a datasheet limit. Above that speed the shear starts to cost reads, and the site gets blamed for a limitation nobody can see. The GT-LPR50M-150 is the model in the range specified for the upper end of that range.
Form factor: what each shape buys and costs
Housing shape changes installation effort and serviceability far more than it changes recognition. The table below is what each shape buys and what it costs, so the choice can be made on the site rather than on the catalogue page.
Two conclusions follow from it. First, for a fixed lane where vehicles pass in one direction, a box camera or an all-in-one unit will both work, and the deciding factor is whether the illuminator can be positioned independently of the lens. Second, a dome is the wrong type for plate work regardless of its resolution, because mounting it on a ceiling or a low soffit puts the plate at a steep downward angle. A PTZ is a reasonable type only where several approaches have to be covered from one pole and traffic is slow, and even then the move has to finish before the vehicle reaches the read zone.
An all-in-one unit is the type that survives contact with a real site best, because it removes the parts that are usually wrong: the lens, the fill light and the processor arrive matched and pre-set, and there is one cable instead of three. The trade-off is that the fill light sits where the housing puts it, so on retro-reflective plates the light angle cannot be tuned separately. That is a capture question rather than a selection one, and the behaviour of reflective film under close illumination is set out in how to test a license plate capture rate on site.
| Type | Where it fits | What it buys | What it costs |
|---|---|---|---|
| Box camera in a housing | Long-range lanes, gantries, multi-lane approaches | Interchangeable lens, separate illuminator, serviceable in place | Two housings, two brackets and exposed cabling; slower to install |
| Bullet | Single lane at 5 to 20 m on a pole or wall | Compact, integrated sunshield, simple to aim | Fixed lens range, and the illuminator position is tied to the housing |
| All-in-one recognition unit | Gates, car parks, tolling lanes, remote sites | Lens, fill light, processor and output in one body; one cable; fewer field settings to get wrong | Fill light position is fixed, which matters on retro-reflective plates |
| Dome | General surveillance coverage | Discreet, wide field of view | Ceiling mounting forces a steep downward angle that flattens the plate; not an ANPR type |
| PTZ | Several approaches from one pole, low speed | Pre-set positions can cover more than one direction | Must refocus and re-expose after every move, and cannot read while travelling |
Housing types compared for plate work
Where recognition runs, and what fails when the link drops
The second axis is where the plate is turned into characters. There are three placements, and they fail differently.
Inside the camera. The unit runs detection, plate location and character recognition on its own processor, at 1 TOPS in the SZGOING range, and outputs a capture image plus structured fields rather than asking a server to watch a stream. The practical consequence is uplink volume. Continuous H.265 at 1080p25 runs from 256 kbps to 8 Mbps depending on scene complexity, and 8 Mbps held all day is about 86 GB per day, while 256 kbps is about 2.8 GB per day. An edge unit that sends records instead of video sits at the bottom of that range, which is what makes a cellular site affordable. The arithmetic is worth doing before choosing a data plan: 86400 seconds multiplied by the bitrate in megabits, divided by eight, gives gigabytes per day.
On a recorder or a server. The camera is treated as a stream source and recognition happens elsewhere. This is the right placement when the recognition engine already exists in the operator's platform and the camera has to feed it, and it is the wrong placement when the site has no cabinet, no air conditioning and nobody to restart a service.
Split. The camera emits the plate record and the business logic lives upstream. This is the common architecture for access control, because the whitelist, the barrier relay and the billing rule belong to the operator, not to the camera, and they change far more often than firmware does.
The choice has one consequence that is easy to overlook: the failure domain. When recognition runs in the camera, a broken uplink costs you live data but not the record, because captures land on local storage. On the 5MP units that is a TF card slot taking up to 128 GB, which at roughly 1 to 2 MB per stored capture is somewhere between sixty and a hundred and twenty thousand records. When recognition runs on a server, a broken uplink stops everything at once, and the site has no plate log until it is restored.
One point of honesty about scope: we manufacture cameras with recognition on board, and we do not sell a separate recognition product for a server. Where an operator needs to integrate with an existing platform, the route is the SDK and the protocol list, which covers ONVIF Profile S into a recorder and TCP, HTTP or MQTT into a platform on Windows or Linux. The ALPR camera model comparison sets out which units carry which interfaces.
Power and backhaul as part of the type
Power and backhaul are usually treated as accessories. They are better treated as part of the camera type, because they eliminate whole categories of site from consideration before any model is discussed.
The solar case is where selection most often goes wrong, and the error is seasonal. A 200 W panel is sized for the worst month at that latitude, not for June. Work out the daily generation from the shortest day, add the cellular router and the fill lights to the load, then allow for several days of low irradiance and for output falling as the panel ages and collects dust. A system sized exactly to the load in midsummer will brown out in midwinter, and the symptom looks like a camera fault rather than a power budget. The BS200-LPR50M-150 solar 4G system is the configuration that pairs the panel, the camera and the cellular link as one unit.
Note that speed class and power class are independent, which is where buyers get caught. A solar unit rated for 150 km/h on a rural highway and a solar unit rated for 50 km/h on a site access road look nearly identical in a photograph, and the second one on the first site will under-read every vehicle that passes at 90. The BS200-LPR40M-50 is the 50 km/h, 30 metre configuration with a 6 to 22 mm lens, and it belongs on an access road, not on a through lane.
| Configuration | Where it fits | What it buys | What it costs |
|---|---|---|---|
| Mains PoE | Gate lanes, tolling lanes, building-mounted approaches | One cable, and the camera sits on the same supply as the switch | Respect the run limit of the PoE standard in use; protect any cable that leaves the building |
| 12 V DC | Legacy poles, long runs, independent supplies | Works where PoE cannot reach | Needs a second cable or a local supply; check voltage drop across the run |
| Cellular, mains powered | Sites with power but no network | No cable to the network, and the uplink is sized to the stream | The data plan has to cover continuous video if live viewing is wanted, not just records |
| Solar 200 W with 4G | Rural roads, temporary sites, construction access roads | No trenching, no mains, no network | Sized on the shortest day of the year, not the longest; panel ageing and dust; fill lights are a night load |
Power and backhaul configurations
Matching type to speed and distance
Speed and distance together narrow the type faster than any other pair of answers, and they narrow it from both ends. Distance sets the lens. Speed sets the shutter. The type has to satisfy both at once.
Two limits in the range are worth stating plainly because they are frequently argued about after the fact. First, the 5MP high-speed models are specified for two lanes per unit. Three lanes of traffic at 1080p means each plate occupies a smaller share of the frame, and the far lane will read worse than the near one no matter how the unit is aimed. Second, the rated speed is a class, not a margin. A unit rated at 50 km/h does not become a 90 km/h unit because the scene is well lit, and a unit rated at 150 km/h is not made better by pointing it at a car park, where the constraint is the downward angle rather than the shutter.
The decision table below is the one to take to a site survey. It assumes a single approach direction and a pole that can be placed where the geometry works; if the pole position is fixed, the distance column becomes the binding constraint and the type is chosen around it.
| Application | Vehicle speed | Working distance | Type | Binding constraint |
|---|---|---|---|---|
| Car park barrier, access lane | Stop to 20 km/h | 4 - 8 m | All-in-one, fixed lens | Plate often angled or close to the frame edge; trigger at the stop line |
| Compound or site gate | Crawl | 5 - 12 m | All-in-one | Tailgating and unplated vehicles; virtual loop plus a no-plate trigger |
| Urban road, up to two lanes | 30 - 60 km/h | 15 - 30 m | Global shutter box or all-in-one | Shutter at 1/1000 s or faster; trigger precision decides the read |
| Highway, tolling, checkpoint | 60 - 150 km/h | 20 - 40 m | Global shutter, 150 km/h rated | 1/2000 s, fill light budget, retro-reflective plate behaviour |
| Rural road with no mains | Up to 50 or 150 km/h by model | Up to 30 or 40 m by model | Solar 4G all-in-one | Winter solar budget; uplink shared with the video stream |
Type by application, speed and working distance
What to put in the RFQ so you get a type, not a range
Selection goes wrong less often from a bad model choice than from a site description that left out the one fact that mattered. These are the nine answers that let a supplier name a type instead of a range, and they are also the nine that a commissioning engineer will ask for a year later when something is wrong.
- Number of lanes and whether traffic in them is one-way or two-way.
- Highest speed vehicles actually reach in the lane, not the posted limit. Take the 85th percentile if there is any data.
- Distance from the likely pole position to the plate plane, and the height available on that pole.
- Whether vehicles stop at a barrier, crawl, or pass without stopping.
- Mains power at the pole, and whether it is on the same supply as the network equipment.
- Network at the pole, and whether a fixed line or cellular is available.
- What the record has to reach: a recorder, a platform, a barrier controller, or a person reading a log.
- Local retention requirement, in days, and whether the captures have to survive an outage.
- Who maintains it, and whether that person can reach the pole, the network port and the configuration interface.
With those nine answers a specification is a short conversation. Without them, the safest advice a supplier can give is a range, and a range is what ends up on the pole.
Related pages
Installation questions we get asked
How can I tell whether an ANPR camera has a global or a rolling shutter?
Park a vehicle with a clean plate at the working distance and check that the characters sit square in the frame. Then freeze a frame of traffic at the speed the site runs and compare a strong vertical stroke, such as a 1 or an H, against the plate border. A stroke that leans while the parked plate was square indicates a rolling shutter. If the strokes stay vertical and the image is soft, the cause is focus or light rather than shutter type.
Do I need a different camera type for 150 km/h than for a car park?
Yes. The car park unit can be an all-in-one with a fixed lens and a rolling shutter, because at walking pace the row-to-row offset is negligible. A 150 km/h lane needs a global shutter, a shutter speed of 1/2000 s, and enough fill light to hold that exposure at night. The two are separate classes in the range rather than two settings on one camera.
Can one ANPR camera cover two lanes?
Two lanes is the specified limit on the 5MP high-speed models. Beyond that, each plate occupies a smaller share of a 1080p frame and the outer lanes lose character detail first, so the capture rate falls unevenly across the road. If three lanes have to be covered, use two units, or mount closer and accept that only the near lanes read reliably.
Is an all-in-one ANPR camera enough without a server?
For capture and plate records, yes. Recognition runs on the 1 TOPS processor inside the unit and captures can be held locally on a TF card up to 128 GB, so a site can produce a plate log with no cabinet at all. What still lives upstream is business logic: whitelists, barrier control, billing and reporting. Those belong to the operator platform and reach it over ONVIF, TCP, HTTP or MQTT.
Which camera type works where there is no mains power?
The solar configurations pair a 200 W panel with the camera and a cellular router. Size the panel on the shortest day at that latitude, add the router and the fill lights to the load, and allow for several days of low irradiance plus output loss from panel ageing and dust. The speed class still has to match the road: a 50 km/h unit on a 90 km/h through lane will under-read regardless of how well the power side is sized.
Does the shape of an ANPR camera affect recognition accuracy?
Only indirectly. Shape decides mounting angle and whether the fill light can be positioned independently of the lens. A dome on a ceiling forces a steep downward angle that flattens the plate, which is why it is the wrong type for plate work. A box camera allows a separate illuminator, which helps on retro-reflective plates. Where the angle and the light are both correct, the housing itself has little effect on the read.
Why does a camera rated at 150 km/h still miss plates on a fast road?
The rating covers the capture chain, not the installation. The usual causes are a trigger point set too early so the plate is small when the frame is taken, an exposure left at the surveillance default, or a shutter that is fast enough per row but still shears across the frame. Each has a different fix, and the diagnosis is quicker if the miss is classified first as a trigger, image or character problem.
Send us the lane and we will name the type
Send us the lane and we will name the type rather than the range: how many lanes, the speed vehicles actually reach, the distance from pole to plate plane, whether there is mains power and a network, and what the plate record has to reach downstream. Those five answers settle the shutter class, the housing and the power configuration before any model number is discussed.
Or email goingtech@sz-going.cn · +86 755 8395 5458