Object Detection: A Comprehensive Review of the State-of-the-Art Methods

The process of localizing and classifying an object in a given sequence of images by computer vision systems is known as Object Detection. The work presented in the area of object detection is categorized into two broad categories. First category of work is based on traditional methods that deals with detection of an object in a single image having no or fewer deformations. The second category of work is based on evolutionary methods that deals with detection of multiple objects in a given image or a sequence of images having deformations. The Evolutionary methods of object detection addresses many core issues like fast detection, multi-view, multi-resolution, object part relation and deformations due to moving object and background. In this work, authors have presented a survey of the state-of-the-art methods of object detection. The object detection methods surveyed in this paper are Histogram of Oriented Gradients based Features, family of Region Proposal based Convolutional Neural Networks, Spatial Pyramid Pooling Network, family of You Only Look Once and Single Shot Detector. This work discusses the methods, training and evaluation aspects of evolutionary object detection methods based on Convolutional Neural Networks and Deep Learning. At the end, open research issues of object detection area are discussed.