A comprehensive review on YOLO versions for object detection

The need for methods used for object detection has gained increasing momentum in recent years. Starting with traditional image processing techniques, this process has been facilitated by the addition of deep learning. Object detection is currently used in areas such as autonomous vehicles, disease diagnosis, robotic vision and industry. The types of systems that are predicted to be needed more and more in the age of developing technology are also increasing. In particular, YOLO (You Only Look Once), which is mostly preferred in real-time object detection, is preferred because it achieves high accuracy in a short time. This paper analyses the main versions of the YOLO algorithm since its first release. The paper systematically analyses the architectural differences between the versions of the YOLO algorithm, the strengths and weaknesses of the models and their contribution to performance. At the same time, in most of the previous studies on YOLO, a comprehensive comparison of the YOLOv9-v11 models is not presented and new architectural features are not evaluated. This review provides an in-depth analysis of the main versions from YOLOv1 to YOLOv11, including recent innovations such as NMS-free, Oriented Bounding Boxes (OBB), GELAN and PGI. This work is intended to be a useful guide for researchers and developers interested in the field.