Explaining and verifying the robustness of Visual Object Trackers to noise

2D tracking is an important computer vision task with important applications in autonomous embedded systems such as Unmanned Aerial Vehicles and autonomous cars that particularly attracted scientists in the past few years. Many new methods have aroused that significantly pushed the state-of-the-art performance in terms of tracking precision, success rate and execution speed, in well-designed and established existing publicly available benchmarks. Despite the fact that these benchmark datasets include as many application scenarios as possible, another commonly neglected yet important aspect is the robustness of tracking methods, notably to noise related with image acquisition, capturing storing and transmission. This paper presents a robustness evaluation toolkit for 2D Visual Object Tracking, that can exploit existing datasets in order to evaluate the robustness of 2D visual tracking methods to realistic image distortion scenarios, mostly encountered in embedded systems. The source code of this toolkit will be made publicly available upon paper acceptance.