Comparative Performance Analysis of YOLOv5 and YOLOv8 for Aerial Surveillance in Smart Traffic Management

Detection and classification of vehicles in various traffic conditions are critical to advances in traffic monitoring and autonomous vehicle technologies. This study addresses the need for more accurate and efficient vehicle detection algorithms by evaluating the performance of YOLOv5 and YOLOv8 object detection models. The smallest and largest submodels (YOLOv5s, YOLOv5x, YOLOv8s, and YOLOv8x) were tested using images captured under extreme conditions, including nighttime and high-altitude drone images. The methodology involves systematically comparing the detection accuracy and inference speed of these models. The results revealed that the ‘x’ submodels outperformed the 's' submodels in terms of detection accuracy across various scenarios. In particular, YOLOv8x demonstrated high detection and classification performance in complex conditions, such as images with varying light filters and dark backgrounds. Conversely, the 's' submodels, although significantly faster, exhibited lower detection accuracy. Additionally, YOLOv8 models outperformed YOLOv5 models across all performance metrics, with YOLOv8x providing the highest accuracy. The study concludes that while there is a trade-off between detection accuracy and inference speed, the YOLOv8x model offers the best balance for applications that require high precision in vehicle detection. These findings provide valuable insights into selecting the appropriate model according to specific application requirements, ultimately contributing to the development of traffic monitoring and autonomous vehicle systems.