Abstract:
|
In this thesis we implemented an automated scanning system for 3D object reconstruction.
This system is composed of a KUKA LWR 4+ arm with Microsoft Kinect cameras placed
on its extreme and thus, in an eye-in-hand con guration.
We implemented the system in ROS using Kinect Fusion software with extra features
added by R. Monica's previous work [16] and MoveIt! ROS libraries [29] to control the
robot movement with motion planning. To connect these nodes, we have coded a suite using
ROS and MATLAB to easily operate them as well as including new features, such as an
original view planner that outperforms the commonly used Next-Best-View planner. This
suite incorporates a Graphical User Interface that allows new users to easily perform the
reconstruction tasks.
The new view planner developed in this work, called Best-Path planner, o ers a new
approach using a modi ed Dijkstra algorithm. Among its bene ts, Best-Path planner o ers
an optimized way to scan the objects preventing the camera to cross again the areas which
have already been scanned. Moreover, viewpoint location and orientation have been studied
in depth in order to obtain the most natural movements and get the best results. For this
reason, this new planner makes the scanning procedure more robust as it assures trajectories
through these optimized viewpoints, so the camera is always looking towards the object
maintaining the optimal sensing distances.
As this project is focused on its later utility in the Intelligent Robotics Laboratory,
we uploaded all the source code in the Aalto GitLab repositories [37] with installation
instructions and user guides to show the di erent features that the suite o ers. |