Realtime Kernel based Machine Learning Template Matching (KMLT)

Abstract

This paper deals with a new approach for the problemof realtime planar templatematching. We consider
tracking as the estimation of a parametric function between observations and motion and we propose an
extension of the learning based approach presented simultaneously by Cootes and al. and by Jurie and
Dhome to non linear regression functions. The estimation of the linear parameters associated to the basis
functions (kernel functions) of the model is then achieved using a training set of motions and associated
observations. We show that the resulting method outperforms the robustness of the linear tracker against
noisy observations.