There’s been some fantastic projects with the Kincet lately, but this is probably one of the most interesting implementations I’ve seen. Kinect’s 3d positioning is used in tandem with an EMF sensor to track and identify the electromagnetic forces around an object in the Kinect’s view, and this data is then rendered and overlaid with the image.
The ready availability of the Kinect’s 3D sensor gives an incredibly accessible method for “digitizing” the real world, to put a somewhat captain power spin on it. Augmented reality applications try desperately to do this via the “lens” metaphor in mobile devices, but suffer from lack of accuracy and lack of relevance.
In this case, the digital is simply permitting an understanding of a physical phenomena that we’ve not the tools to immediately view, shifting the focus from augmentation to assistance. Anyway, really impressed, thanks to Peter Horvath for posting.
As for my own kinect “studies”, I have it working fairly well. My current challenge is rendering the point cloud accurately as a mesh and interpreting movements and intersections in some meaningful way. I’m making use of Toxilibs to handle the really complex stuff spatially, and am trying to use convex hulls to make the transformation mesh wise. Needless to say, a lot of this weekend has been spent staring at Javadocs.
MIT is working on developing the theoretical and practical foundation for machine vision beyond the line of sight context. In short, cameras that can see around goddamn corners.