Real-Time License Plate Recognition Using Edge AI: How Aetina and NVIDIA Jetson Transformed Toll Management in Norway

Key Highlights

• Migrated legacy CPU-based toll systems to GPU-accelerated Edge AI platform in just two weeks using NVIDIA Jetson AGX Xavier
• Reduced power consumption from multi-PC cabinets to approximately 100W per unit while improving real-time object detection and tracking accuracy
• Enabled scalable deployment across Norwegian toll gates and parking facilities with future-ready NVIDIA Jetson AGX Orin architecture delivering up to 275 TOPS

Operational Challenges in Legacy Toll Collection Systems

Norway's traditional toll and parking management infrastructure faced critical limitations that directly impacted revenue collection and operational sustainability. The existing systems relied on outdated fee collection methods and physical barriers that could not reliably ensure comprehensive user payment coverage. This inefficiency compromised funding for essential road and parking area maintenance.

Finter AS, a leading smart mobility technology company operating throughout Norway, identified three major pain points in their infrastructure that demanded modernization through advanced Industrial AI solutions:

Technical and Operational Constraints of CPU-Based Platforms

Finter's existing toll and parking management solution depended entirely on multi-PC and CPU-based processing architectures that generated significant operational challenges. The system produced excessive heat output, consumed substantial electrical power, and required deployment within large, bulky technical cabinets unsuitable for the computational demands of modern machine learning workloads. This CPU-centric approach severely limited Finter's ability to implement sophisticated Computer Vision algorithms necessary for real-time traffic analysis and toll violation detection at scale.

As AI technology matured and advanced deep learning capabilities became industry standard, Finter recognized that their CPU infrastructure could not support the algorithmic complexity required for next-generation Automated Number Plate Recognition (ANPR) systems. The company needed a fundamental architectural transformation to unlock the full potential of Edge AI and GPU-accelerated computing for real-time vision analytics.

Strategic Deployment of NVIDIA Jetson-Powered Edge AI Infrastructure

Finter partnered with Aetina to architect and deploy a complete transition from CPU-dependent systems to NVIDIA GPU-accelerated Edge AI computing. The solution centered on Aetina's advanced edge computing platform built around the NVIDIA Jetson AGX Xavier module, enhanced with M.2 storage expansion and 10G LAN connectivity for seamless integration with existing toll gate infrastructure.

Hardware Architecture and System Integration

Aetina's expertise in Edge AI system integration enabled Finter to complete a comprehensive migration from multi-cabinet CPU setups to a unified, compact ARM-based GPU computing platform within two weeks. This rapid deployment maintained continuous toll collection operations while introducing significant performance improvements. The NVIDIA Jetson architecture provided the foundational computing power necessary to execute real-time Computer Vision algorithms for vehicle detection, tracking, and license plate analysis at every toll gate location.

The physical form factor transformation was equally significant. Aetina's Jetson-based edge devices eliminated the need for large, heat-generating equipment cabinets, enabling streamlined installation at diverse toll gate and parking facility locations across Norway. The compact design reduced installation complexity, logistics costs, and site preparation requirements while improving system reliability through superior thermal management.

AI Software Stack and Algorithm Deployment

Aetina facilitated seamless integration between Finter's custom Computer Vision applications and NVIDIA's comprehensive AI software ecosystem. The deployment leveraged multiple specialized tools from NVIDIA's platform portfolio:

• NVIDIA Metropolis Platform: Provided vision AI application development and deployment framework optimized for toll gate and traffic monitoring use cases
• NVIDIA TAO (Transfer Learning Optimization): Enabled rapid development and customization of vehicle detection and license plate recognition models from pre-trained neural networks
• NVIDIA DeepStream: Powered Displaying Untitled.