Recognition system using Variable Template Network model for meal assistance robot

We propose a system that recognizes positions and shapes of dishes and a spoon on a table, and foods on a dish. First, take input image from above a table; next, dishes, a spoon and foods areas are extracted; finally, each area is matched with Variable Template Network model (VTN model); positions and shapes are recognized. We presented experimental results that our system can recognize rough positions of dishes, a spoon and foods.