An s-pi vision-based tracking system for object manipulation in augmented reality

Abstract

In this paper, we describe a new tracking approach for object handling in Augmented Reality (AR). Our approach improves the standard vision-based tracking system during marker extraction and its detection stage. It transforms a unique tracking pattern into set of vertices which are able to perform interaction such as translate, rotate, and copy. This is based on arobust real-time computer vision algorithm that tracks a paddle that a person uses for input. A paddle pose pattern is constructed in a one-time calibration process and through vertex-based calculation of the camera pose relative to the paddle we can show 3D graphics on top of it. This allows the user to look at virtual objects from different viewing angles in the AR interface and perform 3D object manipulation. This approach was implemented using marker-based tracking to improve the tracking in term of the accuracy and robustness in manipulating 3D objects in real-time. We demonstrate our improved tracking system with a sample Tangible AR application, and describe how the system could be improved in the future.