Comparative Analysis of YOLO-Based Algorithms for Vehicle Detection in Aerial Imagery

Abstract. Real-time object detection has become an essential tool in applications such as traffic surveillance, autonomous vehicles, and industrial monitoring. Among various algorithms, the You Only Look Once (YOLO) series has garnered significant attention for its balance between speed and accuracy. Since its introduction in 2016, YOLO has seen significant advancements and it has been widely adopted due to its ability to provide fast and accurate real-time detection. Over the years, different versions, including YOLO-v1 to YOLO-v11, have introduced improvements in both accuracy and speed. This paper presents a comparative analysis of four recent versions of YOLO-v8-n, YOLO-v9-t, YOLO-v10-n, and YOLO-v11-n focusing on evaluating their detection accuracy and speed in aerial imagery using the EAGLE dataset. Each version incorporates specific advancements aimed at improving performance under different conditions. The study examines the models using a standardized dataset of aerial images with varying illumination and weather conditions. Key performance metrics, such as inference time and Average Precision (AP), are used to evaluate how each model performs in the vehicle detection task in challenging environments. The results provide valuable insights into the suitability of these YOLO models for real-world applications, particularly in dynamic urban environments and areas where traditional camera systems may be less effective. This study aims to identify the fastest and most accurate YOLO model for vehicle detection in aerial imagery using embedded GPU board of Nvidia Jetson AGX Xavier, contributing to the performance enhancement in real-time surveillance and monitoring systems.