Deep Learning based Object Detection Methods: Current Status and Future Prospects

Abstract

Object Detection has evolved significantly since its inception in the early 2000s when Object Localization was regarded as a challenging task. The transition from Localization to object detection has seen rapid advancements in state-of- the-art techniques. The evolution of object detection can be broadly categorized into Traditional Methods and Deep Learning (DL) Methods. In this paper, we examine both approaches, highlighting the rise of YOLO (You Only Look Once) as a dominant force in Object Detection. We provide a comprehensive survey of all YOLO versions, alongside an overview of pre- YOLO object detection methods. Our discussion also explores optimizing object detection through architectural improvements and data enhancements, while considering future opportunities for increased efficiency.