AI-Enhanced License Plate Recognition for Smarter Parking Enforcement
Improving license plate detection and permit validation in challenging real-world conditions.
The project at a glance
ParkSmart's officers scan plates on a mobile app to check for a valid parking permit. We added an AI layer so those scans keep working in snow, mud, glare and motion blur, without changing how officers work.
Plate recognition works in the lab. The curb is a different story.
Snow, mud, glare, damaged plates and motion blur made plates unreadable. Officers had to rescan or verify by hand, and every failed scan slowed the patrol.
Recognition treated every image the same, with no quality check and no clear path for an uncertain read, which risked wrong enforcement decisions.
We started from the curb, not the algorithm.
Dependable reads, fewer rescans and fast permit answers, with no new steps for officers.
Understand the workflow
How officers scan, and what a good answer looks like in the moment.
- Scan flow
- Decision points
- Time pressure
Map the failure modes
Group why reads fail, so each cause can be handled on purpose.
- Weather
- Light and blur
- Angle and damage
Wrap, don't replace
Add an intelligent layer ahead of the engine already in use.
- Lower risk
- Faster to ship
- No retraining
Set production limits
Agree the speed and fallback rules before any code ships.
- Latency budget
- Fallback paths
- Confidence rules
An AI layer that makes every scan smarter.
- It sits between the officer's camera and the existing LPR engine. It judges each image, improves it the way that image needs, then checks how sure the result is.
- Strong reads go straight through, weak ones retry, and unclear ones go to manual review instead of a guess.
- Officers keep the app they know and always get one of three clear outcomes: valid permit, no permit, or unable to determine.
What the product does.
User journey
From scan to enforcement decision in seven steps.
Image Capture
The officer scans a vehicle using the mobile enforcement application. Image quality varies widely depending on lighting, angle and plate condition.
Quality Assessment
The system evaluates the captured image against a set of quality signals — sharpness, lighting, occlusion and angle — to determine how it should be processed.
AI-Assisted Enhancement
Based on the quality assessment, the appropriate image processing pipeline is applied. This may include denoising, contrast enhancement, deblurring or occlusion handling.
LPR Recognition
The enhanced image is passed through the LPR recognition engine, which attempts to extract a plate number with the best possible confidence score.
Confidence Validation
The recognition result is evaluated against confidence thresholds. Low-confidence results can trigger a retry, fallback handling or an escalation path for manual review.
Permit Verification
The recognised plate number is used to query the parking permit database. The system determines whether the vehicle holds a valid, active permit for the location.
Enforcement Decision
The officer receives a clear, immediate result in the mobile application — valid permit, no permit found, or unable to determine — enabling fast, accurate enforcement actions.
How the system fits together
A layered design that adds AI without disturbing the systems already in production.
1. Mobile enforcement app
- Plate capture
- Result display
2. API layer
- Request orchestration
- Built by Tizora
3. AI image processing
- Quality assessment
- Enhancement
- Confidence handling
4. LPR recognition engine
- Plate extraction
- Confidence score
5. Permit database
- Permit lookup
- Validity check
The hard parts, and how we solved them.
Each challenge, the approach we took and what shipped.
Every image is different
Assess first, then pick the fix per image, not one filter for all.
Uncertain reads are risky
Confidence drives thresholds, retries, fallbacks and manual review.
Speed in the field
A latency budget from the mobile workflow, tuned end to end.
Plates that can't be read
Graceful degradation, so enforcement never stalls.
What changed once the AI layer was live.
Qualitative outcomes, before and after.
Fewer repeated scans
Before, difficult plates meant scan after scan. Now a usable read comes far more often on the first try.
Faster, safer decisions
Before, unreadable plates meant manual checks. Now uncertain reads retry or escalate, and permit answers arrive sooner.
Reliable in real conditions
Snow, mud, glare and blur are handled in production, with the same familiar app.
AI delivers the most value when it solves real operational problems — not when it is added simply because it is AI. In this project, we applied computer vision to an existing production workflow where accuracy and reliability directly affect day-to-day enforcement operations.
AI-enhanced parking enforcement FAQ
Related case studies
A fragile scan step became a dependable one.
An AI layer, invisible to officers, turned difficult real-world images into confident enforcement decisions.
