- Version
- Download 14
- File Size 228 KB
- File Count 1
- Create Date October 5, 2021
- Last Updated October 5, 2021
D2.2 Report and Algorithm for the Environmental elements cognition implemented and functional for the knowledge base interface - ABSTRACT
In Sharework robots and human workers cooperate in a shared task. To allow an effective interaction the robot has to understand the intention of the human and the progress of the task. Both are enabled by observing the workspace and detecting items of interest. The environment cognition system fulfills a multitude of high-level requirements. Process items should be detected online and at large ranges. Typical computer vision approaches are limited by either their computational complexity or effective range and accuracy. We aim to find a
tradeoff between accuracy and efficiency to fulfill both characteristics. Therefore, a novel framework is developed to realize modular and customized workspace cognition in many use cases of the manufacturing industry. The main methods developed are an unsupervised, conventional segmentation approach, based on a combination of 2D and 3D vision and a machine-learning classifier capable of detecting particular items in a point cloud. Furthermore, methods are developed and wrapped that allow to calibrate, filter and process networks of cameras of arbitrary size. The algorithms are evaluated in a lab environment using typical workshop items.