Accumulation of different visual feature descriptors in a coherent framework

Abstract

We present a temporal accumulation scheme which disambiguates different kinds of visual 3D descriptors within one coherent framework. The accumulation consists of a twofold process: First, by means of a Bayesian filtering outliers become eliminated and second, the precision of the extracted information becomes enhanced by means of an unscented Kalman filtering process. It is a particular property of our algorithm to be able to deal with different kinds of visual descriptors by the very same mechanism. We show quantitative and qualitative results. © 2011 Springer-Verlag.