Acquiring fuzzy relational model from 3-D hierarchical structure of objects

An approach is proposed that gets a fuzzy relational model of objects by taking 3-D geometrical information of objects obtained from stereo vision as inputs. In order to solve the scale problem occurred in quantitative representation of 3-D structural information of objects and lower the matching cost, a fuzzy relational description language (FRDL) is defined in this paper that represents the 3-D geometrical relations of surface pairs of objects qualitatively. By this FRDL, the quantitative values of 3-D information of objects could be smoothly transformed into qualitative values. This FRDL also serves as the intermediate description between the quantitative geometric values obtained from the vision process and the abstract symbolic description utilized in the machine learning process. With concept learning method, such fuzzy relational model can be generalized and a fuzzy-based 3-D generic model of objects can be learned