Today, I'll be looking at an article from 2009 that discusses an augmented reality application coined "Cultural Heritage Layers". The research team involved in this system had a problem with existing cultural heritage AR projects due to their lack of marker-less tracking systems and doubtful scientific accuracy, as well as the fact that they tend to be built on proprietary (and therefore unsustainable) software. For tackling the first issue, the team built a simple but effective two-step process that the system goes through to learn locations from reference images.
(figure taken from article)
First, randomized trees are constructed based off detectable "key points" of a video frame; these patches are then tracked based on their alignment to the reference image (the tracking is just done via a local search algorithm). This method seems pretty cool to me, although I wish the authors had gone into more detail concerning the feature-detection. They did claim that this method made testing and verification extremely easy, since you could basically just use cardboard cutouts to simulate large buildings.
As far as actually setting up the scene, they relied on a fairly lightweight framework called X3D that just needed minimal OpenGL and 3D acceleration - mostly because they only used 2D textures of historic buildings and scenes, as opposed to actual 3D models. An X3D scene is basically made up of the camera image in the background, a fixed filling quad to display the footage, and a flat ImageTexture node. The idea is that those who wish to set up their own scenes with various options for interactivity can just throw them on top of this node with an extremely small amount of code (the authors claim that only one line of X3D code is needed per point of interest). To demonstrate the system, they briefly described two in-use applications. The first, a visual look at the history of the Berlin Wall, allows users to cycle through historical images by touching the virtual scene on their mobile display. the other, an Italian heritage site called Reggia Vernaria Reale, actually renders the "real" scene differently depending on the image being displayed. For example, a particularly old black and white photo of Italian architecture will draw all people, environments, and real buildings as black and white so that the scene looks seamlessly enhanced with the virtual building.
(figure taken from article)
The work being done here could apply to our system in a few different ways, including the idea of interactivity, 2D image overlays, and tracking. What I really like about this Cultural Heritage Layers project is that it didn't introduce some dazzling new AR algorithm that involved laborious computer vision techniques, but instead focused on existing technologies, an interesting and profitable domain, and usability. Those are the features that AR work needs to improve upon if it is ever to be adopted on a mass scale.
Thanks and gig 'em!
Zoellner, M.; Keil, J.; Drevensek, T.; Wuest, H.; , "Cultural Heritage Layers: Integrating Historic Media in Augmented Reality," Virtual Systems and Multimedia, 2009. VSMM '09. 15th International Conference on , vol., no., pp.193-196, 9-12 Sept. 2009
doi: 10.1109/VSMM.2009.35
URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=5306012&isnumber=5305995
No comments:
Post a Comment