Research Area: Machine Learning
Object detection can be regarded as one of the most fundamental and challenging visual recognition task in computer vision and it has received great attention over the past few decades. Object detection techniques find their application in almost all the spheres of life, most prominent ones being surveillance, autonomous driving, pedestrian detection and so on. The primary focus of visual object detection is to detect objects belonging to certain class targets with absolute localization in a realistic scene or an input image and also to assign each detected instance of an object a predefined class label. Owing to rapid development of deep neural networks, the performance of object detectors has rapidly improved and as a result of this deep learning based detection techniques have been actively studied over the past several years. In this paper we provide a comprehensive survey of latest advances in deep learning based visual object detection. Firstly we have reviewed a large body of recent works in literature and using that we have analyzed traditional and current object detectors. Afterwards and primarily we provide a rigorous overview of backbone architectures for object detection followed by a systematic cover up of current learning strategies. Some popular datasets and metrics used for object detection are analyzed as well. Finally we discuss applications of object detection and provide several future directions to facilitate future research for visual object detection with deep learning.
Keywords:
Author(s) Name: Vipul Sharma, Roohie Naaz Mir
Journal name: Computer Science Review
Conferrence name:
Publisher name: Elsevier
DOI: 10.1016/j.cosrev.2020.100301
Volume Information: Volume 38, November 2020, 100301
