Evidence-based object recognition and pose estimation

The authors present an evidence-based approach to computer-aided-design (CAD)-driven machine vision, in which objects are identified based on salient features found in image data. This approach is successful in identifying and locating objects in range images. This system autonomously learns recognition and pose estimation strategies for objects represented as constructive solid-geometry models, as well as recognition strategies expressed in a boundary representation format. The system performs object recognition with no special manual training or run-time user interaction, and therefore demonstrates the feasibility of true information-driven manufacturing automation