Abstract

This review finds that traditional vehicle tracking systems are time-consuming, error-prone, and ill-equipped to handle real-world complexities like poor illumination and motion blur. The study confirms that a paradigm shift to deep learning (DL) models, especially end-to-end frameworks like YOLO and specialized CNNs, is essential. The key finding is that these DL models internalize the complete process of license plate detection, segmentation, and recognition, which significantly minimizes the cumulative errors and computational latency seen in multi-stage approaches. The review concludes that the integration of resilient Automatic License Plate Recognition (ALPR) and vision-based Speed Violation Detection is paramount for robust Intelligent Transportation Systems (ITS). These advanced systems are crucial for real-time applications in both traffic management (tolling, parking) and law enforcement (speeding, stolen vehicles), ensuring road safety and efficiency with minimal human intervention. Future efforts should prioritize lightweight, edge-optimized architectures.

Key words: deep learning, vehicle tracking, license plate recognition, speed violation detection, intelligent transportation