Date of Award
8-2013
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Legacy Department
Electrical Engineering
Committee Chair/Advisor
Birchfield, Stanley T.
Committee Member
Walker , Ian D.
Committee Member
Gowdy , John N.
Committee Member
Woodard , Damon L.
Abstract
The goal of this research is to develop algorithms using multiple views to automatically recover complete 3D models of articulated objects in unstructured environments and thereby enable a robotic system to facilitate further manipulation of those objects. First, an algorithm called Procrustes-Lo-RANSAC (PLR) is presented. Structure-from-motion techniques are used to capture 3D point cloud models of an articulated object in two different configurations. Procrustes analysis, combined with a locally optimized RANSAC sampling strategy, facilitates a straightforward geometric approach to recovering the joint axes, as well as classifying them automatically as either revolute or prismatic. The algorithm does not require prior knowledge of the object, nor does it make any assumptions about the planarity of the object or scene.
Second, with such a resulting articulated model, a robotic system is then able to manipulate the object either along its joint axes at a specified grasp point in order to exercise its degrees of freedom or move its end effector to a particular position even if the point is not visible in the current view. This is one of the main advantages of the occlusion-aware approach, because the models capture all sides of the object meaning that the robot has knowledge of parts of the object that are not visible in the current view. Experiments with a PUMA 500 robotic arm demonstrate the effectiveness of the approach on a variety of real-world objects containing both revolute and prismatic joints.
Third, we improve the proposed approach by using a RGBD sensor (Microsoft Kinect) that yield a depth value for each pixel immediately by the sensor itself rather than requiring correspondence to establish depth. KinectFusion algorithm is applied to produce a single high-quality, geometrically accurate 3D model from which rigid links of the object are segmented and aligned, allowing the joint axes to be estimated using the geometric approach. The improved algorithm does not require artificial markers attached to objects, yields much denser 3D models and reduces the computation time.
Recommended Citation
Huang, Xiaoxia, "Occlusion-Aware Multi-View Reconstruction of Articulated Objects for Manipulation" (2013). All Dissertations. 1128.
https://tigerprints.clemson.edu/all_dissertations/1128