Robot aided object segmentation based on kinect without prior knowledge

Abstract

As a significant step in artificial intelligence, robot aided object segmentation has drawn increasing attention in past decades. In this paper, a new object segmentation method is proposed based on 3-D information and robot manipulation. We use the Kinect to get the depth information of the object directly as well as the detected contours of it, with which both the primary object segmentation and the 3-D localization of the object can be formed. The usage of Kinect will not only simplify the segmentation process but also improve the performance. Next, the robot arm WAM (Whole Arm Manipulator) with a larger payload is chosen as the manipulator to make further judgment, making it possible to deal with a heavier object. After the manipulation, object segmentation can be accomplished according to the different occasions. Experiments are conducted to prove the effectiveness of this approach.