In this paper we present a new method that robustly identifies doors, cabinets and their respective handles, with special emphasis on extracting useful features from handles to be then manipulated. The novelty of this system relies on the combination of a Convolutional Neural Net (CNN), as a form of reducing the search space, several methods to extract point cloud data and a mobile robot to interact with the objects. The framework consists of the following components: The implementation of a CNN to extract a Region of Interest (ROI) from an image corresponding to a door or cabinet. Several vision based techniques to detect handles inside the ROI and its 3D positioning. A complementary plane segmentation method to differentiate door/cabinet from the handle. An algorithm to fuse both approaches robustly and extract essential information from the handle for robotic grasping (i.e. handle point cloud, door plane model, grasping locations, turning orientation, orthogonal vector to door). A mobile robot for grasping the handle. The system assumes no prior knowledge of the environment.
展开▼
