Visual tracking performance has long been limited by the lack of better appearance models. These models fail either where they tend to change rapidly, like in motion-based tracking, or where accurate information of the object may not be available, like in color camouflage (where background and foreground colors are similar). This paper proposes a robust, adaptive appearance model which works accurately in situations of color camouflage, even in the presence of complex natural objects. The proposed model includes depth as an additional feature in a hierarchical modular neural framework for online object tracking. The model adapts to the confusing appearance by identifying the stable property of depth between the target and the surrounding object(s). The depth complements the existing RGB features in scenarios when RGB features fail to adapt, hence becoming unstable over a long duration of time. The parameters of the model are learned efficiently in the Deep network, which consists of three modules: (1) The spatial attention layer, which discards the majority of the background by selecting a region containing the object of interest; (2) the appearance attention layer, which extracts appearance and spatial information about the tracked object; and (3) the state estimation layer, which enables the framework to predict future object appearance and location. Three different models were trained and tested to analyze the effect of depth along with RGB information. Also, a model is proposed to utilize only depth as a standalone input for tracking purposes. The proposed models were also evaluated in real-time using KinectV2 and showed very promising results. The results of our proposed network structures and their comparison with the state-of-the-art RGB tracking model demonstrate that adding depth significantly improves the accuracy of tracking in a more challenging environment (i.e., cluttered and camouflaged environments). Furthermore, the results of depth-based models showed that depth data can provide enough information for accurate tracking, even without RGB information.

With the development of new methodologies for faster training on datasets, there is a need to provide an in-depth explanation of the workings of such methods. This paper attempts to provide an understanding for one such correlation filter-based tracking technology, Kernelized Correlation Filter (KCF), which uses implicit properties of tracked images (circulant matrices) for training and tracking in real-time. It is unlike deep learning, which is data intensive. KCF uses implicit dynamic properties of the scene and movements of image patches to form an efficient representation based on the circulant structure for further processing, using properties such as diagonalizing in the Fourier domain. The computational efficiency of KCF, which makes it ideal for low-power heterogeneous computational processing technologies, lies in its ability to compute data in high-dimensional feature space without explicitly invoking the computation on this space. Despite its strong practical potential in visual tracking, there is a need for an in-depth critical understanding of the method and its performance, which this paper aims to provide. Here we present a survey of KCF and its method along with an experimental study that highlights its novel approach and some of the future challenges associated with this method through observations on standard performance metrics in an effort to make the algorithm easy to investigate. It further compares the method against the current public benchmarks such as SOTA on OTB-50, VOT-2015, and VOT-2019. We observe that KCF is a simple-to-understand tracking algorithm that does well on popular benchmarks and has potential for further improvement. The paper aims to provide researchers a base for understanding and comparing KCF with other tracking technologies to explore the possibility of an improved KCF tracker.

No abstract

No abstract

Unlike deep learning which requires large training datasets, correlation filter-based trackers like Kernelized Correlation Filter (KCF) use implicit properties of tracked images (circulant structure) for training in real time. Despite their popularity in tracking applications, there exists significant drawbacks of the tracker in cases like occlusions and out-of-view scenarios. This paper attempts to address some of these drawbacks with a novel RGB-D Kernel Correlation tracker in target re-detection. Our target re-detection framework not only re-detects the target in challenging scenarios but also intelligently adapts to avoid any boundary issues. Our results are experimentally evaluated using (a) standard dataset and (b) real time using the Microsoft Kinect V2 sensor. We believe this work will set the basis for improvement in the effectiveness of kernel-based correlation filter trackers and will further the development of a more robust tracker.