System Architecture and Reasoning Scheme in SIGMA
Takashi Matsuyama, Vincent Shang-Shouq Hwang · Sigma · 1990
In this chapter, we first describe the architecture of SIGMA. It consists of the following four reasoning modules: the Geometric Reasoning Expert for spatial reasoning, the Model Selection Expert for appearance model selection, the Low-Level Vision Expert for knowledge-based image segmentation, and the Question and Answer Module for examining the content of the constructed description of the scene. We summarize their roles, knowledge sources, and reasoning methods. Then, we present the principle behind our spatial reasoning, which we call evidence accumulation , and describe how the experts listed above cooperate to construct the scene description. Following several illustrative examples, we discuss the logical foundations of our reasoning method in terms of the first-order predicate calculus. In the last section, we present the knowledge representation scheme for describing the world model: the structures of object models, their relations, and the control knowledge.