一个结合多方面定性空间信息的新方法

A Method of Combining Multi-Aspect Information for Qualitative Spatial Reasoning

定性空间推理是人工智能领域中非常重要的研究内容.空间信息包含拓扑关系、大小关系、形状、距离等很多方面.以往多侧重于单一方面的研究,如何将孤立的各方面信息进行统一表示和推理是当前定性空间推理中的一个重要问题.提出利用结合操作来融合不同空间信息表示的新方法.利用结合操作,可以由原先完备互斥关系集合得到新关系,同时利用原有的复合表自动生成新关系的粗复合表.基于结合操作,给出2个理论模型:结合拓扑关系与大小关系模型、结合拓扑关系与远近关系模型.并提出了邻域划分图的概念,说明了邻域划分图与概念邻域图的关系.利用邻域划分图回答了Galton提出的问题:"为什么LOS(视觉光线演算)的概念邻域图不同于标准的空间或时间关系的概念邻域图,这些关系的复合表中关系总是来自于概念邻域图".

Qualitative spatial reasoning has been an important context in the area of artificial intelligence. Spatial information includes topology, size, shape, distance, etc. Single-aspect spatial information has been studied for many years. But how to combine the single-aspect information in a frame for representation and reasoning is an important problem. In this paper, we propose a new method for combining multi-aspect information using an operation symbol which is called ＂combine＂. By ＂combine＂ operator, one can represent new relations using the single-aspect relation set which is joint exclusive and pair-wise disjoint, and get the rough composition table very easily. Then we give two models. The first one combines the topology and size information and the second one combines the topology and far-near information. We propose a new concept called ＂neighborhood partition graph＂, which could present the relationship among the atom relation in relation set which is joint exclusive and pair-wise disjoint. One can convert the neighborhood partition graph of a new model which combines multi-aspects information into its concept neighborhood graph very easily. We solve the problem proposed by Galton in 1994： ＂why the case of the line-of-sight relations differs interestingly from the standard spatial and temporal relations in that the result of composing two relations does not always form a conceptual neighborhood graph＂.