Visual tracking with structured patch-based model

In this paper, we present a novel structured patch-based visual tracking method, which models the appearance of individual patches and their structural relationships within a unified framework. Specifically, this framework is defined as an optimal patch selection task, and can be further formulated as a linear programming problem, tractable and efficient in tracking scenario. To account for the changing appearance of the target object during tracking process, a pyramid local covariance descriptor is proposed to fuse multiple image characteristics. We compare the proposed method with other competing trackers by the recent large-scale benchmark. Extensive experimental results demonstrate that our tracker performs favorably against the state-of-the-art tracking algorithms.