Comparative Analysis of YOLOv11 and YOLOv12 for AI-Powered Aerial People Detection

YOLO (You Only Look Once) is one of the leading and popular object detection approaches to perform object detection. In this paper a comparison between the latest YOLO models version 11 (YOLOv11) and 12 (YOLOv12) is made for aerial people detection purposes. Here, the experimental analysis is performed in order to observe the main differences and innovations between YOLOv11 and YOLOv12 models delivering state-of-the-art (SOTA) performance. Their performance is assessed using the publicly available VisDrone dataset, specifically designed for visual capturing by drone-mounted cameras. The obtained differences and tradeoffs between CNN (Convolutional Neural Network)-centric architectures and attention-centric architectures are discussed. The goal of this paper is to provide insight into key differences of the training performance and handling of challenging object detection tasks.