Tracking via Robust Multi-task Multi-view Joint Sparse Representation

Consistent multi-layer subtask tracker via hyper-graph regularization

“…Outlier tasks are a set of minority tasks that do not share a common set of features with the majority of tasks. In order to solve the problem, Hong et al [9] further extend the former multi-task learning to multi-task multi-view learning method, where every view in each particle is regarded as an individual task. But the above multi-task learning tracking methods are both constructed based on generative models, and ignore the discriminative information from the background.…”

“…Its extension work [27] further mines the self-similarities among particles via structural multi-task learning to improve tracking performance. With the similar task definition, multi-task multi-view tracking (MTMVT) method [9] is developed to exploit the related information shared between particles and views in order to obtain improved performance. Its robust version is in [3] .…”

An Efficient Misalignment Method for Visual Tracking Based on Sparse Representation

“…Wang [7] regards sparse representation for classification, sampling positive and negative samples, the sparse coefficients obtained in a complete dictionary are used to construct a linear classifier is used to estimate the target candidate's confidence value under two-step particle filtering. Hong [12] treated tracking as a multitask, multi-view sparse learning problem, which utilized multiple views to include various types of visual characteristics, such as intensity, color, and the edges. Each feature can be sparse represented as a linear combination of atom.…”

Pose-Viwepoint Adaptive Object Tracking via Online Learning Approach

“…In [6], Ross et al proposed to incrementally learn a low-dimensional subspace of the target representation. Later, Mei et al [7] introduced sparse representations for tracking, subsequently adopted in many trackers [8,9], in which the memory of the target appearance is modeled using a small set of target instances. In contrast to the generative approaches used in [10] and [11], discriminative methods [12,13,14,15,16,17] have been proposed that consider both foreground and background information.…”

“…Median-Flow tracker [8] uses twice times pyramidal Lucas-Kanade tracker (LKT) [9] to estimate feature points within the last bounding box which sample uniformly in the window, then selects the points whose distance error is smaller than the median distance error and local similarity is bigger than the median similarity. When the number of fitted feature points meets the requirements and median distance error satisfies the thresholds respectively, the tracking result is valid and the tracker outputs the only bounding box.…”