Visual Object Tracking Method’s Analysis and Impact of Deep Learning in Tracking Applications: A Comprehensive Review

Abstract

Visual Tracking is the most elemental research problems in computer vision and finds diversified applications in various fields. This paper aims to introduce various features based tracking methods and analyse it.  Presently, deep learning can classify, detect and recognize the objects with higher accuracy has gained more attraction to the tracking algorithms. Therefore,this paper focuseson deep learning-based methods for object tracking application with consideration of various trackers datasets. The quantitative comparison of deep learning-based tracking algorithms are established using datasets and their attributes and different types of trackers used in deep learning. The analysis proposes that Deep learning outperforms other methodologies with a maximum precision rate of 93.7%. However, due to its supervisory model and iterative nature of deep learning algorithms, it is computationally complex and demands lots of computational power. This is a major drawback of deep learning to use it in real-time applications. Hence, superior hardware is required for faster computation.It has been observed that it can achieve a maximum speed of 100 frames / second in offline tracking by avoiding the fine-tuning.