The You Only Look Once (YOLO) series has become the most prominent framework for real-time object detection, offering an exceptional balance between speed, accuracy, and efficiency. Since the introduction of YOLOv1, each successive version has progressively addressed core challenges such as multi-scale detection, computational efficiency, and generalization in diverse environments. However, persistent limitations remain in areas such as small object detection, deployment on edge devices, and robustness under occlusion or complex scenes. This paper presents a comprehensive survey of the YOLO family, from YOLOv1 to the recent YOLOv12, detailing the architectural innovations, training strategies, and performance improvements that define each generation. We categorize these developments across key components, including backbones, detection heads, label assignment strategies, and data augmentation pipelines. Special attention is given to the transition from anchor-based to anchor-free designs and the integration of transformer-based modules in modern versions. Our comparative analysis on COCO and Pascal VOC benchmarks reveals YOLOv11-L offers the optimal speed–accuracy balance among large models. Finally, we discuss remaining challenges and future directions, including the adoption of neural architecture search (NAS), domain adaptation, and scalable deployment for real-world applications. This study aims to serve as a foundational reference for researchers and practitioners aiming to design accurate and efficient real-time object detectors.