Performance Analysis on Deep Learning State of Art Algorithms for Object Recognition

Abstract

The goal of computer vision, a subfield of computer science, is to replicate some of the intricacies of the human visual system so that machines can recognize and interpret images and videos in the same manner that humans do. Until recently, computer vision was only used in a restricted capacity. In the past few years, artificial intelligence has advanced significantly, outperforming humans in a number of tasks involving object detection, recognition, and classification. This has allowed computer vision to grow exponentially in terms of increasing the precision with which machines can recognize the objects in and around the surrounding environment. A computer vision technology called object recognition helps find and identify objects in a series of images and videos. Despite the fact that the image of the things varies in different viewpoints, different sizes and scales, or when they are translated or rotated, humans can recognise a large number of objects in images with minimal effort. Even when partially obscured from view, human vision system has the greatest capability to identify the objects. Whereas, for computer vision systems, this task is still a difficulty. Over the years, several different approaches and innovations in the algorithm have been tried to impose the human’s capability into a computer’s vision system. This paper provides a thorough investigation on the evolution of Object Recognition algorithms, datasets used and its performance metrics in a precise manner which will guide the future researchers a direction to proceed their research in innovating algorithms with better accuracy.