Decentralized Edge-AI System for Real-Time Syrian License Plate Recognition Using Khadas Vim3

The growing dependence on personal vehicles has intensified challenges in traffic management, violation monitoring, and locating stolen cars. Addressing these issues, this study introduces a real-time, smart system for detecting and recognizing Syrian car license plates using surveillance camera image processing. The system is built on a decentralized architecture with edge computers, notably the Khadas Vim3 equipped with Neural Processing Units (NPUs). It utilizes two YoloV4-tiny networks trained on a dataset of over 1100 images of Syrian license plate. The first network detects cars and plates with high accuracy, while the second recognizes numbers on the plates with 87% accuracy. Adhering to Internet of Things (IoT) principles, the system processes up to 16 frames per second for detection and 8 frames per second for broadcasting results, maintaining uninterrupted operation even if an edge computer is offline, thereby providing a consistent and efficient solution for traffic management challenges.