Improved Object Tracking via Bags of Affine Subspaces

A robust visual tracking system requires the object model to handle occlusions, deformations, as well as variations in pose and illumination. To handle such challenges, in this paper we propose a tracking approach where the object is modelled as a continuously updated set of affine subspaces, with each subspace constructed from the object's appearance over several consecutive frames. Furthermore, during the search for the object's location in a new frame, we propose to represent the candidate regions also as affine subspaces, by including the immediate tracking history over several frames. Distances between affine subspaces from the object model and candidate regions are obtained by exploiting the non-Euclidean geometry of Grassmann manifolds. Quantitative evaluations on challenging image sequences indicate that the proposed approach obtains considerably better performance than several recent state-of-the-art methods such as Tracking-Learning-Detection and Multiple Instance Learning Tracking.