The Evolution and Advancement of YOLO Algorithms in Object Detection: From Real-Time Breakthroughs to Modern Architectures

Object detection represents a foundational capability in Artificial Intelligence (AI), enabling machines to interpret visual environments through precise object localization and classification. This comprehensive review chronicles the revolutionary evolution of the You Only Look Once (YOLO) framework from its inception to the state-of-the-art YOLOv12. Beginning with the limitations of classical approaches using handcrafted features, YOLO’s paradigm-shifting is documented transition to unified real-time detection via regression-based architectures. Methodically analyzing each major version (v1- v12), key innovations is detailed including multi-scale predictions (v2/v3), anchor-free designs (v8), programmable gradient information (v9), and attention-enhanced cross-scale fusion (v12). The review establishes how successive iterations systematically addressed critical challenges: reducing computational latency by 47× versus R-CNN variants, improving mAP by 32.7% on COCO benchmarks, and enabling deployment on edge devices. Beyond architectural analysis, comparative performance evaluations is presented across diverse applications—from autonomous driving to medical imaging—demonstrating YOLO’s unprecedented balance of speed (142 FPS) and accuracy (78.4% AP). The paper further examines emerging implementation trends, hardware optimizations, and domain-specific adaptations that cement YOLO’s position as the de facto framework for real-time vision systems. Our review analysis provides both technical and historical context for researchers and practitioners navigating the landscape of modern object detection.