Interfaces for people with severe motor impairments that enables them to interact with a healthy
person and manipulates a wheelchair by extending the body motion ability through information equipment
recently has been put to practical use. It is well known that tongue and gaze motion can remain even in
much serious stage. It is a problem to assure the electric insulation of the device in oral cavity. In this
research, we propose a pointing I/F using the intraoral image taken by a small camera for the endoscope
available at low cost. Since the individual movement of the tongue is largely different according to disease
condition and it may changes over time, a learning signal to the CNN is dynamically generated and adapted
to ensure the invariant operative output range.