Date of Award
2026
Degree Type
Thesis
Degree Name
Master of Science in Mechanical Engineering (MSME)
Department
Mechanical, Industrial and Systems Engineering
First Advisor
Musa Jouaneh
Abstract
In automated scanning for industrial refinishing applications, large workpieces often exceed the reach of a single robot base position, and may require the environment to be scanned from multiple locations and stitched to a common reference frame. This thesis compares two robotic scanning workflows for such an object: a full scan that reconstructs the object from a single robot base position, and a segmented scan occurring across multiple simulated robot base positions that are stitched together through fiducial-based localization. The results from both workflows are compared for measurement accuracy of the object's bounding box against known dimensions, position of the bounding box relative to a world coordinate system against known physical reference measurements, and overall processing time.
A second output of this thesis is to provide a fully functional pipeline to setup both scanning workflows for different equipment configurations. The main equipment used in this thesis is an OAK-D pro camera for RGB-D imaging and an Epson VT6 robotic arm to manipulate the camera. Python is the main coding language used, with an ordered sequence of setup and execution scripts designed to reproduce results under different conditions. The pipeline includes: camera calibration to obtain the intrinsic matrix, distortion coefficients, and reprojection error; hand-eye calibration using a customizable hemispherical toolpath for varied viewing angles; multi-marker ArUco localization with breadth-first-search (BFS) marker-graph chaining and targeted drift correction to reduce localization noise; iterative closest point (ICP) registration for point cloud stitching, with tunable methods and parameters; and object measurement via plane segmentation and constrained bounding-box extraction.
The results indicate that the full scan and segmented scanning workflows produce comparable results for measurements, localization accuracy, and processing time. In general, measurement accuracy was more dependent on processing parameters or scanning configurations, and not on the full scan vs segmented scan workflow. This suggests that ArUco-based localization combined with ICP refinement allows a globally stitched, multi-position point cloud to match the accuracy of a point cloud stitched entirely from a single robot base position.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Egan, Casey, "INVESTIGATION OF AUTOMATED 3D SCANNING STRATEGIES FOR LARGE-SCALE POLISHING APPLICATIONS" (2026). Open Access Master's Theses. Paper 2756.
https://digitalcommons.uri.edu/theses/2756