Range-data analysis guided by a junction dictionary

摘要

A knowledge-guided system for the analysis of range data about three-dimensional scenes with planar and/or curved objects is presented. Inputs to the system are range data originated with real scenes, and outputs are the concise descriptions (i.e., organized structures of vertices, edges and faces) of objects, which are useful for object identification and machine manipulation. A “junction dictionary” is constructed to represent the general knowledge about the physical nature of the three-dimensional objects world, and the system uses the dictionary for the feature extraction and the feature organization. At each step of the analysis the system consults the dictionary to predict positions, orientations and physical types of missing edges. Those predictions enable the system to decide where and what kind of edges to search for as well as to decide how to organize the extracted features into a reasonable description. Experiments show satisfactory system performance.