Home / Projects / PlatePatrol
Project 02 / 04 — Active

PlatePatrol

Built by Exact IOT — internal product

A mobile ANPR app that turns any smartphone into a portable stolen vehicle detection system. Using on-device AI, PlatePatrol reads license plates from the live camera feed and instantly checks them against a locally-synced copy of the NZ Police stolen vehicle database. All processing happens on-device — no cloud, no uploads, no privacy concerns.

Real-time HW-accelerated Scanning
6 On-device ML Models
0 Data Sent to Cloud
PlatePatrol scan screen showing a matched plate with WATCH badge and live camera feed
The Challenge

No practical way to spot stolen vehicles

Identifying stolen vehicles on the road is largely passive and reactive. Members of the public have no practical way to check whether a vehicle they see is on the stolen list. The NZ Police publish their stolen vehicle database publicly, but it's a raw CSV download — not something you can use while walking past a car park.

Existing commercial ANPR systems are expensive, fixed-installation hardware designed for police and parking enforcement. They require dedicated cameras, backend servers, and ongoing subscriptions that make them impractical for individuals or community groups. There was no affordable, portable tool that could do real-time plate recognition using hardware people already carry in their pocket.

PlatePatrol database screen showing synced NZ Police stolen vehicle data with search
The Solution

On-device AI that runs without the cloud

PlatePatrol runs multiple on-device AI models to detect and read license plates from the live camera feed in real time. The app uses hardware-accelerated neural processing (NPU/GPU) when available, delivering the best experience on modern flagship phones (2021+) with dedicated AI silicon — though it runs on any supported device with graceful CPU fallback.

The app automatically syncs the NZ Police stolen vehicle list and stores it locally. When a scanned plate matches, the app triggers an audio alert and saves the detection with a photo, GPS coordinates, and timestamp. Unmatched plates are never persisted — a deliberate privacy-first design. The app also supports manual photo capture, configurable accuracy-vs-speed tradeoffs, and tap-to-focus with pinch-to-zoom for real-world usability.

PlatePatrol detection details showing STOLEN badge, full frame photo, confidence score and vehicle info
App Screenshots

See it in action

Scan screen — matched plate with WATCH alert

Matched plate alert

Detection details with STOLEN badge and vehicle info

Detection details

Database screen with synced stolen vehicle list

Stolen vehicle database

Scan screen in normal mode with clean plate detected

Normal scanning mode

Technical Details

Under the hood

PlatePatrol is built with .NET MAUI, shipping a single C# codebase to both Android and iOS. Six ONNX models run on-device via hardware-accelerated neural processing — no data ever leaves the phone. The app automatically detects and uses the best available compute on each device: dedicated NPU, GPU, or CPU fallback. Unmatched plates are never written to disk, and matched detections auto-delete after 30 days.

C# .NET MAUI ONNX Runtime SQLite
platepatrol.specs
Platforms Android + iOS
Language C# / .NET 10
AI Processing On-device, HW accelerated
Data Source NZ Police DB (auto-synced)
Local Storage SQLite + 30-day cleanup
Privacy Unmatched plates never stored
Billing Play Store + App Store subs
Like What You See?

Need a custom mobile
AI solution?

We build on-device machine learning applications that run without cloud dependencies — from concept through to app store deployment.

Start a Conversation Back to Home