Monocular 3D vision for a robot assembly environment.

摘要

The three-dimensional position and orientation of objects is estimated from one or more monocular images. The identities of the objects are known, and geometric models are assumed to be available. Images are interpreted, finding correspondence between linear features extracted from sensor data and modeled object features, by searching an interpretation tree built using prior position estimates. Object positions are updated by maximum-likelihood estimation. Position estimation results from an implemented system are presented, demonstrating location of partially occluded objects in a cluttered scene. 16 refs.