Rapid invocation and matching of 3D models to 2D images usingcurvilinear data

摘要

Describes a practical approach to match 2D scenes to 3D models inwhich the scene data is described as a set of curvilinear featuresextracted from a grey scale image of a rigid object taken from anarbitrary viewpoint, and the model data is derived from a CSG modeller.In contrast to systems that directly acquire depth data, the low andintermediate processing is more reliable and less complex, which canlead to a more effective system using a rapid hypothesis andverification cycle. This may change as research in 3D increasinglyconcentrates on improving the sensor technology and low-levelsegmentation algorithms so that surface descriptions become lessfragmented and include plausible curved surfaces