Measurement Data Recording and Preprocessing for Training Data Generation using ROS

The ongoing development of automated machine control systems amplifies the necessity of large-scale data acquisition and
storage. Data collection and preprocessing have a major impact on the quality of the developed systems and therefore have to be optimized and monitored. Especially for machine learning systems preprocessing is a key factor to assure functionality. Our measurement setups are based on ROS and connect a variety of sensors and data sources to one network. The data collection was thereby automated with a variety of developed software tools that feature a simple graphical user interface for quality monitoring and reducing the effort for the operator during the data collection process. In this paper, we present the developed data collection tools and the preprocessing tools to convert the data into training data for neural networks to help other researchers in their data collection tasks.