Confidence measures for variational optic flow methods

Abstract

In this paper we investigate the usefulness of confidence measures for variational optic flow computation. To this end we discuss the frequently used sparsification strategy based on the image gradient. Its drawbacks motivate us to propose a novel, energy-based confidence measure that is parameter-free and applicable to the entire class of energy minimising optic flow techniques. Experimental evaluations show that this confidence measure leads to excellent results, independently of the image sequence or the underlying variational approach.