Back to AI Solutions
Computer Vision

AI-Powered CCTV &
License Plate Recognition

Real-time vehicle detection and number plate reading using YOLOv9c — trained on Australian plates using a hybrid Apple Silicon MLX and NVIDIA GPU pipeline. Integrates with your existing cameras — no cloud required.

90.9%
Plate Accuracy
<5s
Alert Delivery
24/7
Continuous Monitoring
0
Cloud Dependencies

How It Works

Your existing CCTV cameras stream video to a local AI processing unit. The system detects vehicles in real time, reads license plates, logs every entry, and sends instant alerts when watchlist matches are found — all without any data leaving your premises.

CCTV and license plate recognition data flow pipeline showing cameras, AI processing, vehicle detection, plate reading, database logging, and instant alerts
1

Video Capture

Your existing IP cameras stream RTSP video to the local Frigate NVR. Any brand, any model — if it streams video, it works.

2

AI Detection

YOLOv9c runs on local GPU hardware detecting vehicles and isolating license plates in real time. Faster and more accurate than its predecessor — no internet connection needed.

3

Read & Alert

Plates are read with OCR, logged to a searchable database, and checked against your watchlist. Matches trigger instant phone alerts.

Technical Details

YOLOv9c Frigate NVR NVIDIA CUDA TensorRT Apple MLX Tesseract OCR RTSP Home Assistant Pushover Alerts
Model & Training Details

Detection Model: YOLOv9c (You Only Look Once) — the latest YOLO architecture with improved accuracy and efficiency — trained on custom Australian vehicle and plate datasets using a hybrid Apple Silicon MLX and NVIDIA RTX 3090 pipeline.

OCR Engine: Tesseract with Australian plate format preprocessing for standard, custom, and interstate plates.

Inference Speed: 10-25ms per frame on RTX 3090, enabling real-time multi-camera processing.

Training Data: Thousands of Australian plates across varying conditions — day, night, rain, angles, speeds.

Training Pipeline: Apple Silicon (M-series) with MLX handles rapid prototyping, data augmentation, and model experimentation. Full training runs and TensorRT optimisation are performed on NVIDIA RTX 3090 (24GB VRAM).

Hardware Requirements

Recommended GPU: NVIDIA RTX 3090 (24GB VRAM) or newer. RTX 4090 for maximum throughput.

Camera Capacity: 8-16 simultaneous streams per GPU depending on resolution.

Storage: 1-4TB SSD recommended for 30-90 day event retention.

Network: Gigabit LAN between cameras and processing unit. No internet required.

Who This Is For

Commercial Properties

Car park management, tenant access control, delivery vehicle logging for office buildings and shopping centres.

Residential Estates

Gated community access, visitor tracking, unknown vehicle alerts for luxury homes and estates.

Warehouses & Logistics

Fleet tracking, delivery verification, unauthorised vehicle detection for distribution centres.

Body Corporate

Apartment complex parking enforcement, resident vs visitor tracking, after-hours monitoring.

Frequently Asked Questions

Does AI CCTV work with my existing cameras?
Yes. Our system works with any IP camera that supports RTSP streaming, which covers most modern security cameras including Hikvision, Dahua, Reolink, UniFi Protect, and many others. We integrate with your existing infrastructure — no need to replace cameras.
How accurate is the license plate recognition on Australian plates?
Our YOLOv9c model achieves 90.9% accuracy on Australian license plates in production conditions. This includes various lighting conditions, angles, and speeds. The model is specifically trained on Australian plate formats including standard, custom, and interstate variations.
Does the CCTV footage leave my premises?
No. All video processing happens locally on GPU hardware installed at your site. Video streams never leave your network. The system runs entirely on-premise with zero cloud dependency, making it suitable for high-security environments.
What hardware do I need for AI CCTV analytics?
For production inference we recommend an NVIDIA RTX 3090 or newer GPU, which handles 8-16 camera streams simultaneously depending on resolution and frame rate. Our training pipeline also uses Apple Silicon with the MLX framework for rapid model development and experimentation before full training runs on the RTX 3090. We handle all hardware specification and setup as part of the project.
Can I set up alerts for specific vehicles?
Yes. You can maintain watchlists of specific license plates and receive instant push notifications when a match is detected. Alerts are delivered to your phone within seconds of detection. You can also set up time-based rules, such as alerting only outside business hours.

Ready to Make Your Cameras Intelligent?

Turn your existing CCTV into an AI-powered security and analytics system. Local processing, instant alerts, zero cloud fees.

Book a Free Consultation Call 0400 819 487