Visual Object Tracking with Deep Learning

In its simplest definition, the problem of visual object tracking consists in making a computer recognize and localize persistently a target object in a video. This is a core problem in the field of computer vision that aims to replicate the human ability in keeping the focus on a particular object with the sight. In the past, several different algorithmic principles have been proposed to reach such a capability. Thanks to the tremendous improvement in accuracy, recent algorithms based on deep learning emerged as promising methodologies to achieve the goal. The fundamental idea behind these techniques is to exploit the ability of deep neural networks in learning complex functions to learn how to track objects by visual examples. The potential of this kind of tool attracted the interest of the research community so much that nowadays deep learning is the way-to-go for the implementation of effective visual tracking algorithms. Despite the popularity, the study of deep neural networks for visual tracking is relatively at its early stages. This means that there are still many open issues that need to be addressed to fully comprehend the capabilities and potentialities of such learning models. In this Thesis, we try to give an answer to some of these questions.