Lessons Learned from Human Pose Interaction in an Industrial Spatial Augmented Reality Application

Abstract

This paper is a technical description of a novel Augmented Reality application in the industrial domain of furniture production. In the presented case, workers suffered from high cognitive load in doing end-of-line quality inspection and individual handling of a high variety of products. The proposed solution consists of a Spatial Augmented Reality system, where a projector directly displays information on the product to assist the worker. At the same time results of an in-line quality inspections are shown which can be interactively modified through human gestures.The main contribution of this work is two-fold: (1) in contrast to other industrial augmented reality applications the described one technically builds upon a deep neural net based pose estimation. This allows a seamless interaction with the system and tracking of human actions rather than deriving them from the current state of the work piece, (2) the paper recapitulates experiences and results of the approach with a focus on lessons learned for using human pose estimation in smart production.