Vtrack: a novel visual object tracking benchmark database

Abstract

Visual object tracking is one of the most emerging areas in the field of computer vision. In recent years, significant progress has been achieved in this research area. Recent research in this field is mainly focused on data acquisition for creating new benchmark databases suitable for evaluating the performance of various tracking applications, designing several tracking methods, etc.

The detection ability of any tracking method depends largely on the database on which it is trained. In order to perform efficient training of tracking methods and to properly evaluate their tracking capability, suitable benchmark databases containing videos or image frames with various attributes—e.g., occlusions, blurriness, etc, are required.

The work conducted in this thesis is mainly focused on performing effective benchmarking of tracking capabilities of multiple methods. The main contribution of this thesis lies in designing a novel database, namely VTrack: a visual object tracking benchmark database comprising 25 videos containing thousands of frames with various attributes.