Accurate Estimation of Motion Vectors using Active Block Matching

Abstract

We propose an active block matching algorithm for motion estimation. The proposed algorithm dynamically determines the search area and the matching metric. We exploit the constraint of small velocity changes of a block along the time to determine the origin of the search area. The range of the search area is adjusted according to the motion coherency of spatially neighboring blocks. Our matching metric includes multiple features. The degree of overall match is computed as the weighted sum of matches of individual features. We adjust the weights depending on the distinctiveness of features in a block, so that we may discriminate features according to the characteristics of an involved block. The experimental results show that the proposed algorithm can yield very accurate block motion vectors.