Dataset Generation Using a Simulated World

Abstract

In this paper we focus on a missing part of solving the detection of human workers around the load and the truck of a crane. Current solution attempts use Convolutional Neural Networks to detect these workers in images. Especially the detection around the load can not be solved with the use of public datasets as the viewing angles differ to much. We therefore propose and evaluate an approach which uses a simulation engine to create automatically labeled images to train a network for this use case.