PoseNetwork: Pipeline for the Automated Generation of Synthetic Training Data and CNN for Object Detection, Segmentation, and Orientation Estimation

Abstract

The latest developments and research of convolutional neuronal networks (CNNs) have proven the feasibility of their use in industrial applications that require object detection and pose estimation in unknown environments. Nevertheless, the end-users have neither the required resources for model-training nor the expertise to efficiently implement such applications. On the one hand, our work proposes a pipeline that focuses on the automated generation of training data by using synthetic images. On the other hand, we introduce a deep neural network to estimate the orientation of a reference object by using a one-shot image. We demonstrate the use of PoseNetwork by detecting and estimating the 5D-Pose of a workpiece in a robot-based inspection cell.