Learning structural concept with 3-D information of objects
G. Dong, Takuro Yamaguchi, M. Yachida · 2002
A new approach is proposed which learns structural concepts using learning from example, by taking 3D information of objects obtained from stereo vision as input for the system. In order to solve the scale problem in the quantitative representation of 3D information, the concept description language (CDL) is defined which represents the 3D relations of surface pairs of objects qualitatively. This CDL representation also serves as the intermediate description between the quantitative values obtained from the vision process and the abstract symbolic description utilized in the machine learning process.>