Efficient Learning of Linear Predictors for Template Tracking

The research on tracking templates or image patches in a sequence of images has been largely dominated by energy-minimization-based methods. However, since its introduction in Jurie and Dhome (IEEE Trans Pattern Anal Mach Intell, 2002), the learning-based approach called linear predictors has proven to be an efficient and reliable alternative for template tracking, demonstrating superior tracking speed and robustness. But, their time intensive learning procedure prevented their use in applications where online learning is essential. Indeed, Holzer et al. (Adaptive linear predictors for real-time tracking, 2010) presented an iterative method to learn linear predictors; but it starts with a small template that makes it unstable at the beginning. Therefore, we propose three methods for highly efficient learning of full-sized linear predictors—where the first one is based on dimensionality reduction using the discrete cosine transform; the second is based on an efficient reformulation of the learning equations; and, the third is a combination of both. They show different characteristics with respect to learning time and tracking robustness, which makes them suitable for different scenarios.