Abstract
Understanding the layout of objects in a room should be a strong assistance for detecting free space in which a robot can move. As object recognition, we propose a method to use the general shape concept of object, an example of which for a chair is "it has a seating surface at certain height", "it may have a back surface", etc. Using 3D point cloud data captured by a range image sensor, successive process are performed, which are segmenting regions, extracting planes, and recognizing object categories based on abovementioned conceptual rules. It was experimentally proved that our method can perform the extraction and recognition of desks, chairs, and small objects such as pet-bottle on the desk. Demonstration using a mobile manipulator was carried out to verify the effectiveness of proposed recognition method.
