Abstract
In this paper, we proposed a novel visual tracking system by constructing the Constrained Correlation Filter (CCF) with Depth Information. More specifically, in our proposed system, to avoid the boundary effects and the fixed shape assumption of conventional Discriminative Correlation Filter (DCF), the shape of target is extracted from depth image provided by RGB-D sensor to construct the CCF, which may prevent the filter from being disturbed by the background noise at the learning stage and enlarge the search region. Moreover, in order to avoid the drifting problem, the update of the model is stopped once part of the target is occluded. The feature weighting coefficients, which reflect the discriminability of the feature channels, are used at the location stage to improve the discriminability. The experimental results show that our method is capable of achieving state-of-the-art performance on Princeton RGB-D tracking benchmark among all public tracking algorithms.
Export citation and abstract BibTeX RIS
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.