Probabilistic Pose Recovery Using Learned Hierarchical Object Models

Renaud Detry (1) , Nicolas Pugeault (2) , Justus Piater (1)

1 Université de Liège
2 SDU - University of Southern Denmark

This paper presents a probabilistic representation for 3D objects, and details the mechanism of inferring the pose of real-world objects from vision. Our object model has the form of a hierarchy of increasingly expressive 3D features, and represents 3D relations between these probabilistically. Features at the bottom of the hierarchy are bound to local perceptions. While we currently only use visual features, our method can in principle incorporate features from diverse modalities within a coherent framework. Model instances are detected using a Nonparametric Belief Propagation algorithm which propagates evidence through the hierarchy to infer globally consistent poses for every feature of the model. We present an importance-sampling mechanism for belief updates that is critical for efficient and precise propagation. We finally present a series of pose estimation experiments on real objects, along with quantitative performance evaluation.

Mots clés: Computer vision; 3D object representation; pose estimation; Nonparametric Belief Propagation.; Nonparametric Belief Propagation

Domaines: Apprentissage [cs.LG]

Renaud Detry, Nicolas Pugeault, Justus Piater. Probabilistic Pose Recovery Using Learned Hierarchical Object Models. Journées Francophone sur les Réseaux Bayésiens, May 2008, Lyon, France. ⟨hal-00259480v2⟩