A REVIEW ON OBJECT DETECTION SINCE 2005

Abstract

Recent years have seen a significant increase in interest in object detection, which is considered to be one of the most fundamental and demanding computer vision tasks. It might be called the pinnacle of computer vision history, because of its rapid progress since 2005. If we regard today's object identification to be a kind of deep learning-powered technical aesthetics, then rewinding the clock since 2005 would allow us to observe the wisdom of the cold war period. This study examines studies on object detection since 2005 in the context of technological advancements that have occurred throughout a quarter-century. This article covers a wide range of subjects, including historical milestone detectors, metrics, datasets, speed-up strategies, etc. This article also reviews important traditional object detections (DPM and HOG), single-stage object detection methods (YOLOR, RetinaNet, SSD, YOLO), and two-stage object detection methods (Mask-RCNN, Faster-RCNN, Fast-RCNN, SPPNet, R-CNN).