Contour/texture approach for visual tracking

Abstract

In this article, the problem of real-time hybrid contour/texture tracking for planar objects is addressed. On one hand, a lot of methods have been proposed to track objects from their contours. On the other hand, numerous other tracking algorithms deal with texture. In real situations, objects can unfortunately rarely be divided so clearly. Therefore, an hybrid tracking approach, able to mix texture and contour information, appears to be very useful. The proposed approach is very simple and efficient. It is based on image differences, which are the differences between object aspects in the image and aspects predicted using a parametric transformation model. Knowing a difference image, the proposed algorithm only need a matrix multiplication to estimate motion parameters. This is possible due to the use of an off-line learning stage. © Springer-Verlag 2003.