Johannes Rauer
Mohamed Aburaia
Wilfried Wöber
2020-09-20
<p>Machine-learning-based approaches for<br>
pose estimation are trained using annotated groundtruth<br>
data – images showing the object and information<br>
of its pose. In this work an approach to semiautomatically<br>
generate 6D pose-annotated data, using<br>
a movable marker and an articulated robot, is<br>
presented. A neural network for pose estimation is<br>
trained using datasets varying in size and type. The<br>
evaluation shows that small datasets recorded in the<br>
target domain and supplemented with augmented images<br>
lead to more robust results than larger synthetic<br>
datasets. The results demonstrate that a mobile manipulator<br>
using the proposed pose-estimation system<br>
could be deployed in real-life logistics applications<br>
to increase the level of automation.</p>
https://doi.org/10.5281/zenodo.4084909
oai:zenodo.org:4084909
Zenodo
https://zenodo.org/communities/fhtw
https://doi.org/10.5281/zenodo.4084908
info:eu-repo/semantics/openAccess
Creative Commons Attribution 4.0 International
https://creativecommons.org/licenses/by/4.0/legalcode
Semi-Automatic Generation of Training Data for Neural Networks for 6D Pose Estimation and Robotic Grasping
info:eu-repo/semantics/article