Occlusion Handling in Augmented Reality User Interfaces for Robotic Systems

摘要

When working with Augmented Reality (AR) interfaces, achieving realistic occlusion between real and virtual objects can be critical for certain applications: it is one of the most decisive elements in depth perception of humans. Several techniques for achieving accurate and stable occlusion with moderate computational effort are presented in this paper: first, a model-based approach is shown, that is extended to include the measurements of a spatial sensor for also determining dynamic occlusions with objects that are not in the existing 3D-model. Finally, a technique that does not require any predefined 3D-models at all is introduced using distance information from a 3D time-of-flight camera. This approach also allows leaving the egocentric view of the camera that is providing the pictures for the AR-application to an arbitrary exocentric view, and can be directly employed on a stereo display without actually deploying a real stereo camera.