抄録
The authors have been studying a direct and intuitive 3-D geometric operation interface using a deformable real object as an input tool. To use various-shaped input tools other than parallelepiped ones as in our previous reports, this paper proposes a neural network learning algorithm to optimize number of units in the input layer as well as in the intermediate layers. The algorithm enables us to initially put surplus number of electric terminals to various-shaped block of elastic-conductive material to measure its deformation and then optimize the number by removing useless terminals. Also, we extend free form deformation so that is can be applicable to various (non-parallelepiped) shape by intermediately using a simplified control volume approximating the objective control volume by some parallelepipeds. The proposed ideas are implemented as a prototype interface system, and the efficacy and possibility of our approach are confirmed.
