A condominium in Ho Chi Minh City wants to replace paper tickets at its vehicle gate. The developer has quotes for a barrier system, an RFID tag scheme, and a license plate recognition camera. All three open the gate. Only one of them can tell you how long each vehicle stayed, whether the same car returned three times this week, and which units are parking overnight without authorisation — and that difference is why ANPR projects are usually justified on data rather than on the barrier itself.
This guide covers how ANPR works, where it fits, what drives accuracy, and the questions to settle before installation.
What ANPR does, and what it replaces
Automatic number plate recognition (ANPR), also called license plate recognition (LPR), reads the characters on a vehicle's plate from a camera image and converts them into a searchable record: plate string, timestamp, camera location, and usually a snapshot.
In a parking or access-control context, that record replaces several manual steps at once:
- A driver no longer takes a ticket or holds a card to a reader.
- Entry and exit are matched automatically, so duration and fee calculation become a lookup.
- Monthly and season-pass vehicles are recognized and admitted without staff intervention.
- A searchable history exists for investigations, disputes, and occupancy reporting.
RFID and ticket systems still win in some cases — heavy rain, extreme angles, or plates that are frequently obscured by trailers or dirt. The honest comparison is not "which technology is better" but "which failure mode can this site tolerate."
How the recognition pipeline works
Every ANPR system, whatever the vendor, runs the same sequence:
- Detection. The camera or analytics engine finds a vehicle in the frame.
- Localization. It isolates the plate region within the vehicle image.
- Character recognition. Optical character recognition converts the plate characters into text, and often a country or region classification.
- Matching and action. The system compares the plate against a list — permitted, denied, season pass, blacklist — and triggers a barrier, a notification, or a record.
Steps 2 and 3 are where accuracy is won or lost. Everything before that is trigger logic; everything after is integration.
Where ANPR fits
Paid parking
The core use case. Entry and exit cameras capture plates, the system pairs them, and fee calculation happens on duration. The business case usually rests on reduced staffing, fewer ticket-related disputes, and better occupancy data rather than on the plate reader alone.
Gated communities and staff access
Residents, staff, and frequent visitors can be placed on an allow list. Unknown vehicles are held for guard confirmation or a visitor workflow. In practice, the guard workflow matters more than the recognition itself — a system that admits residents automatically but has no clean way to handle visitors will be worked around within a week.
Retail curbside and loading areas
Retailers use plate capture for curbside pickup verification and to manage loading bays. Here the value is a log: which vehicle, which bay, how long. Accuracy requirements are lower because a human usually confirms the plate at the counter.
Logistics yards and fleet control
Fleet and logistics sites combine plate reading with a yard management system to track arrivals and departures. Integration quality decides whether this works; standalone recognition adds little.
What drives accuracy
Plate recognition fails for physical reasons before it fails for software reasons. The factors that matter most:
- Camera angle. A plate is designed to be read by a human standing in front of it. Cameras mounted at steep vertical angles or extreme horizontal offsets have to reconstruct characters from a distorted image, and accuracy drops.
- Illumination. Headlights, night conditions, and strong backlight are the classic failure modes. Infrared illumination solves part of it but produces monochrome images, which changes how reflective and retroreflective plates appear.
- Vehicle speed and motion blur. A fast-moving vehicle at a slow shutter speed produces a smear. Matching shutter speed to expected approach speed is a design decision, not an afterthought.
- Plate condition and type. Dirty, damaged, tilted, or non-standard plates are read less reliably. So are plates on trailers, motorcycles, and vehicles with tow hitches.
- Plate formats across Southeast Asia. Formats, character sets, and plate sizes vary by country, and some markets use scripts that are harder for generic models to read. Country-specific training and testing matter more than a global accuracy claim.
- Weather. Heavy rain and glare degrade every optical system.
The practical rule: expect accuracy figures quoted by vendors to describe a well-mounted camera under good conditions. Design the workflow so that a misread is recoverable — a guard confirmation step, an image shown to the driver, or a manual lookup — rather than treating recognition as infallible.
Integration: the part that decides the project
Recognition is a data source; the value comes from what consumes it.
- Barriers and gates. Most integrations use dry-contact relay outputs or a software interface to the controller. Confirm which, and how the system behaves when a vehicle is admitted while another is still on the loop.
- Access-control platforms. Where a site already runs an access-control system, the plate reader usually feeds it as another credential type.
- Video management systems. ANPR events should appear alongside footage so an operator can review a denied vehicle immediately.
- APIs and databases. Ask whether plate events can be pushed to your own system, in what format, and at what latency. A closed appliance with no export is a dead end for multi-site reporting.
- Tailgating and anti-passback. Plate recognition does not stop a second vehicle following the first through a barrier. If that matters, it is a separate design problem.
Test the integration with the actual barrier model on site before the rollout. Integration surprises are the most common cause of ANPR projects running late.
Data protection: settle this before installation
A license plate identifies a vehicle and, in many cases, an identifiable person — which means plate data is generally treated as personal data. The specific rules vary by market, and the details matter more than the general principle.
Regimes you will encounter in the region include the GDPR in Europe, Singapore's PDPA, Malaysia's PDPA, Thailand's PDPA, the Philippines' Data Privacy Act, Indonesia's personal data protection law, and Vietnam's Decree 13 on personal data protection. Each imposes its own requirements on collection, notice, retention, and access; some treat biometric and identification data more strictly than others.
Practical steps that hold up in most of these regimes:
- Post clear signage at the point of capture, in the local language.
- Define a retention period per purpose and delete on schedule.
- Restrict who can search plate history, and log those searches.
- Avoid exporting plate data to personal devices.
- Decide in advance how a subject access request would be handled.
Because the requirements differ by country and can change, treat this section as a starting point and confirm the exact obligations for your deployment with legal counsel.
What to specify in an RFQ
- Purpose of each lane (entry, exit, or both) and expected vehicle types.
- Approach speed range and plate mounting height range.
- Country and plate formats in scope, and whether mixed-format reading is required.
- Lighting conditions at each lane, including night and backlight.
- Recognition performance on a sample set you provide, not a generic claim.
- Integration points: barrier model, access-control platform, VMS, and API.
- Retention period, storage location, and whether analytics data stays on site.
- Commissioning plan with a two-week acceptance period.
- Support terms, spare parts, and firmware update policy.
Insist on the sample-set test. A vendor who will run your own plate images through their engine and report results is giving you more information than any datasheet.
FAQ
Can ANPR replace a parking ticket system entirely?
For most passenger-car parking, yes, provided the workflow handles exceptions: misread plates, motorcycles without front plates, foreign plates, and drivers with no matching entry record. Keep a manual override at every barrier. In heavy rain, at awkward angles, or where plates are often obscured, a hybrid approach is more realistic.
How accurate is license plate recognition?
Accuracy depends on the camera position, illumination, vehicle speed, and plate condition at that specific lane. Well-designed lanes do better than difficult ones, and quoted figures rarely reflect a bad angle at night in the rain. Ask for performance on a sample of images from the actual location.
Does ANPR work at night?
Yes, usually with infrared illumination or adequate visible lighting, but night accuracy is lower than daytime accuracy and both must be tested separately. Confirm how the camera behaves with headlight glare on a wet road.
Can one camera cover two lanes?
Sometimes, if the geometry allows each vehicle to be captured at a workable angle. In practice, one camera per lane with a defined capture zone produces more predictable results, and multi-lane single-camera designs need an on-site demonstration before you commit.
Is plate data personal data?
In many jurisdictions it is treated as personal data because it can be linked to an identifiable person. The obligations differ by country. Confirm the requirements for your market with legal counsel before go-live, and design retention and access controls accordingly.
---
Vehicle recognition is usually one part of a wider project, and the solutions overview shows how it sits alongside video surveillance, access control, and people-flow analytics. The product center lists the camera families HooViz builds for parking and perimeter deployments. To discuss a specific lane layout or gate configuration, contact our team.