The AugmenTable is an augmented reality (AR) workstation that we use to investigate natural visual perception for
monocular AR displays. An interface is described as natural if its technical realization is effectively veiled from the user and when it emulates the typical appearance of the presented items. It is well known that the natural perception enhances spatial understanding of scenes. Common AR workstations that are in use in industry present a monoscopic video of physical fixtures or parts, on a large 2D screen or monitor. Thus, the users lose the advantages of natural perception.
The goal of this research is a method to enable natural visual perception for monocular AR workstations that facilitates intuitive assembly of mechanical systems. The novel solution is the utilization of a hemispherical image obtained from a hemispherical mirror that provides a 180º viewing angle. Computer vision techniques are used to simulate monocular depth cues such as perspective changes, motion parallax, and shadows. Our hypothesis: depth cues allow to better estimate the distance and sizes of objects, in general, to better understand the spatial structure of objects.
The main hardware setup is a tabletop with a monitor mounted on one side of the table. A 6-inch diameter reflective hemisphere is attached to the back of the monitor. The hemisphere reflects the entire environment, including the working area, and provides an omnidirectional image of the environment. A video camera stands on the opposite side of the table. It captures images of the tabletop via the reflective sphere. A second video camera, referred to as head tracking camera, is attached on top of the monitor. It observes the user, particularly the movements of his/her head.
To improve natural visual perception, we simulate the monocular depth cues perspective, motion parallax, and shadows. The starting point is the video image that is obtained from hemisphere as well as the user's head position. Perspective is simulated by selecting a subset of the entire 180º image that aligns with the user's head position. Motion parallax is simulated by identifying objects in the scene and by moving the objects with respect to the user's head position and their distance to the user. This is implemented with a layered image technique. A shadow texture technique is used to render shadows on virtual and physical objects. In summary, by simulating depth cues, the user gains the impression to observe a natural 3D scene.
A user study has been conducted to evaluate the efficiency of simulated depth cues. Our hypotheses are that a user is able to better estimate distances and sizes when depth cues are simulated. Thus, users have been asked to estimate the sizes of simulated and virtual objects as well as their distance to the user. This has been compared with a regular AR application. The results indicate that simulated depth cues improve the estimation of distances.
For further information, results have been published in:
R. Radkowski, J. Oliver, 2014, "Enhanced Natural Visual Perception for Augmented Reality-Workstations by Simulation of Perspective," IEEE Journal of Display Technology, 10(5), May 2014, pp. 333-344
R. Radkowski, J. Oliver, 2014, "Monocular Depth Cues for Augmented Reality Applications to Enhance Spatial Perception Tasks," I. Horváth and Z. Rusák (Eds.), Proceedings of Tools and Methods of Competitive Engineering (TMCE 2014), May 19–23, Budapest, Hungary
R. Radkowski, J. Oliver, 2013, "Simulation of Motion Parallax for Monitor-based Augmented Reality Applications," Proceedings of the ASME 2013 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference, IDETC/CIE 2013, August 30-September 2, Portland, OR
The Augmented Reality Lab explores the augmented reality (AR) technology and its capabilities for engineering applications.
© 2014 All Rights Reserved