The rapid development of intelligent transportation systems and connected vehicles has led to the generation of massive volumes of diverse and heterogeneous traffic data. Efficient analysis, interpretation, and classification of this data are crucial for enhancing mobility, traffic prediction, and safety in modern transportation networks. Recent studies have demonstrated that deep learning models are capable of effectively capturing both spatial and temporal dependencies in traffic datasets, enabling more accurate and reliable analysis and classification compared to traditional methods. However, most existing approaches focus solely on software-based implementations, often overlooking the practical challenges of deploying these models in real-time, resource-constrained embedded environments.

This review provides a comprehensive analysis of deep learning approaches applied to vehicular traffic data classification. It emphasizes the evaluation of model effectiveness, computational efficiency, and suitability for implementation in embedded systems, highlighting various optimization and adaptation strategies that make deployment feasible in hardware-constrained contexts.

The study also highlights current research trends, identifies critical open challenges in achieving real-time inference on limited-resource hardware, and discusses potential future directions for integrating deep learning methods with embedded systems. By bridging the gap between deep learning model design and practical hardware implementation, this review contributes to the development of intelligent, efficient, and deployable AI solutions for next-generation connected and autonomous vehicles, ultimately supporting safer and more effective transportation networks.