On the Use of Multi-Depth-Camera Based Motion Tracking Systems in Production Planning Environments

Abstract

In order to keep up with changing market demands like highly individualized products, automotive OEMs are facing these new requirements with mass-customization and more efficient digital methods. With the ongoing virtualization of production planning and verification, new technologies are necessary to facilitate interaction with virtual environments for planning experts. While marker-based motion capture systems have been used for tracking of worker movements and interaction with those digital models until now, with the advent of markerless tracking technologies like the Kinect depth cameras, those solutions attract more interest because of their reduced cost, ease of use and the absence of cumbersome suits and markers. However, when using depth cameras, which have been primarily designed as short-range gaming devices, different issues arise when dealing with large-scale setups common in shop floor and production planning environments. Similar to traditional motion capture systems, it is necessary to deal with occlusions, camera registration and calibration and fusion of multi-depth-camera data to obtain a sufficiently large and robust tracking space. For this purpose, we propose such a novel multi-depth-camera approach, which is able to register itself only by observing the worker's trajectory. It is then able to reliably track the worker movements across a large space and also optimize the skeletal tracking. Through its distributed and scalable approach, the multi-depth camera system can be set up flexible for versatile tracking scenarios in practical production planning use. Based on sensor hardware specific quality criteria, it is able to fuse incoming tracking data and to provide a coherent view of the tracking space even in difficult tracking situations.