Detection of Moving Objects for Robotic Manipulation Applications in Non-structured Environments

Robotic manipulation tasks, particularly in unstructured environments, require detailed information about the manipulator surroundings. Location and form of objects to be manipulated are especially important information. Due to the emergence of RGB-D cameras, e.g. the Microsoft Kinect, the research community w as able to develop important results related to 3D environment mapping, scene segmentation and object detection. Calibration between two cameras, depth camera and RGB camera provided in the RGB-D devices has been theme of an extensive number of papers. Nevertheless not only the intrinsic and extrinsic cameras parameters are necessary, the transformation between the reference frames of the camera and the reference frame of the manipulator is mandatory information for any arbitrary application. Frequently this measurement procedure is difficult to perform and implies in parameter values deviations. This paper presents an experimental method to obtain the transformation parameters for the two reference frames. The robot modeling is elegantly developed using the product of exponentials approach and the twist between robot and camera reference frames which is obtained, and represents the best solution in a least squares sense. Experimental application using a SCARA 4 D.O.F robot and a Kinect RGB-D camera is conducted to ascertain the method effectiveness. This approach represents an improvement and a flexibilization for robot programming tasks.