3D object modeling and recognition via online hierarchical Pitman-yor process mixture learning

We present a statistical framework for 3D objects modeling and recognition. Our framework is based on describing 3D objects using local descriptors from which a visual vocabulary if built and on a hierarchical Pitman-Yor process mixture of Beta-Liouville distributions. An online approach based on variational Bayes is developed for the learning of the proposed framework. The merits of our model are shown via extensive experiments.