Robust Manipulability-Centric Object Detection in Time-of-Flight Camera Point Clouds

This paper presents a method for robustly identifying the manipulability of objects in a scene based on the capabilities of the manipulator. The method uses a directed histogram search of a time-of-flight camera generated 3D point cloud that exploits the logical connection between objects and the respective supporting surface to facilitate scene segmentation. Once segmented the points above the supporting surface are searched, again with a directed histogram, and potentially manipulatable objects identified. Finally, the manipulatable objects in the scene are identified as those from the potential objects set that are within the manipulators capabilities. It is shown empirically that the method robustly detects the supporting surface with ±15mm accuracy and successfully discriminates between graspable and non-graspable objects in cluttered and complex scenes.