Object tracking via appearance modeling and sparse representation

This paper proposes a robust tracking method by the combination of appearance modeling and sparse representation. In this method, the appearance of an object is modeled by multiple linear subspaces. Then within the sparse representation framework, we construct a similarity measure to evaluate the distance between a target candidate and the learned appearance model. Finally, tracking is achieved by Bayesian inference, in which a particle filter is used to estimate the target state sequentially over time. With the tracking result, the learned appearance model will be updated adaptively. The combination of appearance modeling and sparse representation makes our tracking algorithm robust to most of possible target variations due to illumination changes, pose changes, deformations and occlusions. Theoretic analysis and experiments compared with state-of-the-art methods demonstrate the effectivity of the proposed algorithm.

Graphical abstractIn this paper, we propose a tracking algorithm based on the combination of appearance modeling and sparse representation (Fig. 1 shows the graphical abstract of the proposed tracking algorithm). Although the appearance manifold of an object may be quite nonlinear and complex, we make a reasonable assumption that the manifold can be approximated by multiple linear models. With this assumption, we learn multiple linear subspaces to model the target appearance variations during tracking. In order to make our tracking algorithm more robust to abrupt target appearance changes due to occlusion or image corruption, we combine the learned appearance model and the sparse representation method to further learn a similarity measure for distinguishing the target object from background. More specifically, when abrupt appearance variation occurs, the target appearance can be reconstructed by the learned appearance model and a limited number of noise bases. Tracking is then achieved by Bayesian inference, in which a particle filter is adopted to estimate the target state sequentially. With the tracking results in new frames, we update the appearance model adaptively. Compared to other tracking methods, the proposed method shows two improvements in dealing with appearance variations of the target. First, different kinds of target observations can be modeled by different subspaces of the constructed appearance model. When the target shows a large appearance change, new subspaces will be added to cover it. Second, sparse representation is introduced to deal with abrupt appearance changes in the proposed method. When the target undergoes partial occlusions, the learned appearance model can still represent it well due to the additional noise bases.Figure optionsDownload full-size imageDownload high-quality image (127 K)Download as PowerPoint slideHighlights► We model correlated observations of an object by a linear subspace. ► We use multiple subspaces to model the appearance manifold of an object. ► We combine appearance modeling and sparse representation to deal with variations.