Real-Time Vehicle Detection Using Surveillance Cameras: An Empirical Evaluation in Vietnamese Traffic Scenes

Real-time vehicle detection from surveillance cameras is essential for traffic management, safety, and urban planning. The complex and dynamic nature of Vietnamese traffic, characterized by congestion, varied vehicle types, and frequent occlusions, poses significant challenges that necessitate a robust evaluation of vehicle detection mod-e<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$l$</tex>s. YOLO models are renowned for their high detection accuracy and fast processing speeds, making them suitable for real-time applications. However, as these models evolve, their increasing complexity and size may hinder their effectiveness, particularly on resource-constrained devices. This study rigorously assesses the efficiency of these YOLO models in real-world Vietnamese traffic scenarios, analyzing metrics such as detection accuracy and processing speed to identify specific shortcomings and areas for improvement. We evaluates the performance of various YOLO models (v5 to vlO) on the UIT- VinaDEVES22 dataset, specifically designed for Vietnamese traffic scenes. Our findings reveal that while YOLO models demonstrate strong performance in vehicle detection, their growing size may limit their practicality in real-time applications. This empirical evaluation provides valuable insights for optimizing vehicle detection systems, contributing to the advancement of computer vision in challeneing traffic environments.