Week 12
Marker - Based Video Tracking
Most video tracking methods have a significant margin of error, particularly when you want them to work in general purpose conditinos. One way to get more precise input is to use a system that recognizes visual markers. These are usually high-contrast patterns printed onto objects that the system can recognize using edge-detection. This approach can be very accurate and can give not just position but also orientation and sometimes even position in 3D space.
ReactiVision is a computer vision engine that tracks objects with special patterns (called Fiducial Markers) applied to them. You can print out the image and glue it to anything. The ReactiVision software is a standalone app that connects to other programs over Open Sound Control, making it very flexible and also able to run on its own computer. It's open-source and cross-platform.
Another even more powerful (though more complicated) system is AR Toolkit (Augmented Reality Toolkit) which can precisely determine position and orientation of objects in 3 dimensions. You can hook it into VRML, OpenSceneGraph, or OpenSG as an input into interactive 3D VR worlds.
And I should mention OpenCV which is one of the most powerful computer vision systems available. Also an open-source project, it's a C/C++ library with more advanced capabilities like feature recognition and object differentiation.