Three dimensional Bayesian state estimation using shearlet edge analysis and detection

In this work, we present a method of estimating the kinematic state of a three dimensional object from a set of image sequences recorded at multiple views. In our approach, three dimensional information from a Bayesian filter is merged to stabilize two dimensional recognition as well as tracking so observation collection and object state estimation are concurrent. A unique aspect of this method is that a shearlet transform is used to reliably extract image features. The method is demonstrated on both synthetic and real data for performance evaluation.