A NOVEL MULTILINEAR PREDICTOR FOR FAST VISUAL TRACKING

This letter presents a novel prediction scheme employed for fast visual tracking. The proposed multilinear predictor is formulated as a higher order tensor, instead of the existing vector representations. This predictor is based on emploing the Canonical/Parallel factors (CP) decomposition to decompose a tensor as a sum of rank one tensors. In that way, the proposed scheme efficiently retains the underlying structural information of the input data, while reduces at the same time the compu-tational complexity by employing separable filter operations applied at different directions. The effi-ciency of the proposed scheme is demonstrated in the conducted experiments.