The most common complaint in license plate recognition projects is the same: it reads by day but not at night; it reads cars but not lorries. The problem almost never starts in the software. The camera is in the wrong spot, the plate is too small in the image, the shutter is too slow, and the IR setting turns the plate into a white blob. LPR (License Plate Recognition) or ANPR (Automatic Number Plate Recognition) is analytics that finds the plate in the image and converts its characters to text; it’s used at factory and logistics entrances, corporate car parks and campuses for entry–exit logs, barrier control with an allow list and shipment tracking. Its success depends more on camera position, optics and image settings than on the software.
Plate pixel width
LPR software needs the plate to be a certain width in the image; manufacturers generally recommend values in the 100–250 pixel range, depending on the model. The standard long plate in Türkiye is about 52 cm wide. Calculation:
Plate pixel width ≈ horizontal resolution × (plate width ÷ field-of-view width)
If a camera with 1920 pixels of horizontal resolution sees a width of 4 meters at the distance where the plate is: 1920 × 0.52 / 4 ≈ 250 pixels. If the same camera sees 8 meters, ≈ 125 pixels. That’s why reading two lanes with a single camera is often not recommended.
Do the calculation at the point where the plate will be read, not at the farthest point the camera sees. As the vehicle approaches the barrier, the plate passes through the image over several frames; the software needs to capture enough pixel width in at least a few of these frames. Putting the read zone where the vehicle slows down or stops means a higher read rate with the same camera.
Angle: horizontal and vertical
The camera should see the plate as head-on as possible. As horizontal and vertical angles grow, characters slant and distort; manufacturers generally recommend angles below 30 degrees. Camera height, the point where vehicles stop or slow down, and vehicle types (cars, trucks, lorries) are assessed together; a lorry’s plate and a car’s plate are at different heights.
A common mistake on site is mounting the camera on the gatehouse roof or an existing lighting pole “because the cable is already there”. An easy mounting point often means the wrong angle. At entrances with mixed vehicle traffic, if a single camera can’t read both plate heights at the same quality, planning separate cameras or separate lanes for cars and heavy vehicles is sounder. On lorries, the front plate is often under the bumper and exposed to dirt; at some sites, reading the rear plate gives more reliable results.
Shutter and motion blur
A fast shutter is needed for the plate to be sharp on a moving vehicle. General-purpose cameras gain brightness in low light by slowing the shutter, which creates motion blur. On LPR-specific cameras or profiles, the shutter is fixed (for example 1/500 s or faster, depending on vehicle speed) and the missing light is made up with IR.
If you’re using a general-purpose camera for LPR, turn off auto shutter in the night profile and keep WDR and digital noise reduction (DNR) low. High DNR softens character edges. An image that looks cleaner on screen can be worse for reading.
Night and headlight glare
Turkish plates have a retroreflective background; they reflect IR light back. With correctly set IR and shutter, headlight glare is suppressed and the plate characters appear sharp. With the wrong settings, the plate becomes a white blob. That’s why a night acceptance test is mandatory.
Two situations need attention in the night test: reading while the vehicle’s headlights point directly at the camera, and the plate being lit too closely by the camera with excessive IR. Smart IR or profiles that adjust IR power to distance reduce the second problem.
Software and plate format
- Region/country setting support for the Turkish plate format and font
- Two-line plates, old-style and special plates
- Reading vehicles with foreign plates (especially at lorry entrances)
- Read confidence score and threshold setting
If the country setting is wrong or left in multi-country mode, the software hesitates between O and 0, I and 1, B and 8. If an allow-list matching rule tolerates such similar characters, the tolerance should be enabled only in search and reporting; an exact match should be required for the barrier decision.
How is accuracy measured?
| Measurement | Method |
|---|---|
| Capture rate | For how many passing vehicles was a plate record created? |
| Exact read rate | How many read plates match the real plate character for character? |
| Partial reads | Reads with one or two wrong characters (may be useful in search, not in a barrier decision) |
| Breakdown by condition | Day/night, rain, dirty plates, high speed |
| Latency | Time from the vehicle to the barrier command |
Measurement is done by comparing with a manual log of vehicles passing over a set period, or with general camera footage. Results should be reported separately for day and night.
Example calculation: 200 vehicles passed during the night, a record was created for 190, and 176 of those records were correct character for character. The capture rate is 190/200 = 95%; the exact read rate relative to passing vehicles is 176/200 = 88%. For barrier automation, the second number is the meaningful one. In the acceptance report, state clearly which rate is calculated against what; “95% read rate” on its own says nothing.
Barrier automation and process
Automatic barrier opening shouldn’t go live before the read rate is accepted. For unreadable vehicles or those not on the list, a fallback process is defined, such as attendant approval, an intercom or an access card. Pre-registration and time-limited allow lists can be used for visitor vehicles. Integration options are described on the access control integration and car park solutions pages.
Design errors happen on the barrier side too. Without a vehicle detection loop or barrier safety sensor, a barrier that opens when a plate is read can close while the vehicle behind is passing. The camera triggers the barrier to open; closing safety is the barrier system’s job. The source of the allow list must also be clear: is the list fed automatically from HR or the visitor system, or updated by hand at the security desk? If a departed employee’s vehicle stays on the list, the problem isn’t the camera.
The personal data dimension
Since plate data can be linked to the vehicle owner, it may count as personal data. The purpose (entry control, shipment logging), retention period, access permissions and privacy notice should be set by the organization. On the technical side, access to plate records is restricted by role and automatic deletion is applied at the end of the period. If the plate database and video recordings have different retention periods, check that each is configured separately; even if video is deleted, the plate table can pile up for years.
Frequently asked questions
How many megapixels does a license plate recognition camera need?
Megapixels alone don’t decide it; what matters is how many pixels wide the plate is in the image. For a narrow lane, 2 MP may be enough, while reading two lanes with a single camera requires higher resolution or a narrow lens.
Are dirty and damaged plates read?
The read rate drops. In acceptance testing, such plates are reported in a separate category, and a fallback procedure (approval by an attendant) is defined for automatic barrier decisions.
How long should we keep plate records?
Since plates can be linked to the vehicle owner, they may count as personal data. The period should be set by the organization according to purpose and obligations, and expired records should be destroyed.
We carry out the study and measured installation for your facility entrances as part of our license plate recognition service.