Tracking a planar patch by additive image registration

Abstract

We present a procedure for tracking a planar patch based on a precompiled Jacobian of the target region to be tracked and the sum of squared differences between the image of the patch in the current position and a previously stored image of if. The procedure presented improves previous tracking algorithms for planar patches in that we use a minimal parameterisation for the motion model. In the paper, after a brief presentation of the incremental alignment paradigm for tracking, we present the motion model, the procedure to estimate the image Jacobian and, finally, an experiment in which we compare the gain in accuracy of the new tracker compared to previous approaches to solve the same problem. © Springer-Verlag Berlin Heidelberg 2003.