A comprehensive survey on object detection YOLO
This review covers the evolutionary journey of YOLO from its initial release to the latest versions, encompassing an in-depth analysis of the performance and critical characteristics exhibited by each iteration.
Abstract
As a single-stage object detection framework, the YOLO (You Only Look Once) technique has emerged as a prominent technique for various object detection tasks owing to its impressive balance between speed and precision. This research article presents a comprehensive review of the YOLO family of algorithms. This review covers the evolutionary journey of YOLO from its initial release to the latest versions, encompassing an in-depth analysis of the performance and critical characteristics exhibited by each iteration. Particular emphasis is given to exploring the applications of YOLO in diverse domains, focusing on its role in real-time object detection on embedded systems. Furthermore, the paper delves into the latest advancements in compressing algorithms for optimizing the cumbersome YOLO models and practical implementation examples. The potential of deploying YOLO on resource-constrained devices is further unlocked by addressing the challenge of model size reduction. Finally, this study outlines potential research trends and improvements for the YOLO family of algorithms, including novel architectural designs and innovative training strategies. Overall, the thorough investigation presented in this review is a valuable reference for researchers seeking to explore the YOLO framework and its evolving landscape in object detection.