In this thesis, the motion parameter estimation algorithm based on the quadrantal division of moving images is proposed. Six parameters are used, which can describe any arbitrary two-dimensional motion of rigid objects.
In the first step, a change detector distinguishes between temporally changed and unchanged regions of two successive frames. The motion estimation algorithm assumes that a uniform motion occurs within a connected changed region. In each changed region, one set of motion parameters is estimated. If each changed region is not correctly described by its estimated mapping parameters, the mapping parameters, after splitting image into 4 subimages, are estimated again for each subimage with changed region. This process is continuously performed by splitting the subimages with changed region into 4 smaller subimages and estimating mapping parameter until each changed region is described well by one mapping parameter set. In next step, adjacent regions with the same motion are merged into a new larger region.
Experimental results show that the proposed estimatation technique is able to split a changed region which has several motion into different subregions which are described by one mapping parameter set each.