Recognition of overlapped objects using RGB-D sensor

To realize a practical object recognition system, we employ RGB-D sensor to treat overlapped objects. In this paper, we focus on recognition of partially overlapped breads. Appearance of upper side breads often deforms its shape in captured images, which degrades recognition accuracy. On the other hand, lower side object is partially hidden by other breads, so we cannot achieve whole appearance in captured images. In the proposed method, in order to improve the recognition accuracy of overlapped objects, we adjust or complement the shape of breads by consulting depth information achieved by depth sensors. Some experimental results show that by adjusting overlapped object shape and by complementing hidden area, we can improve recognition accuracy of overlapping objects.