A stereoscopic machine-vision system offers a capability for robust tracking of objects in three dimensions (3D). The system is suitable for use in a variety of robotic applications. The system hardware includes a pan-tilt-verge camera and aiming mechanism assembly and a board that holds control and image data processing circuitry. Included on the circuit board are circuitry and firmware that perform correlations at the speeds necessary for tracking complex objects in real time. The system software implements algorithms for correlation-based tracking of objects, using stereoscopy. A significant portion of development effort was directed toward solving the problem of gaze control — that is, where to focus attention and vision image data processing resources. The chosen solution lies in the stereo-cluster approach, which embodies tenets of active vision. In somewhat oversimplified terms, a stereo cluster can be characterized as a cluster of points, selected on the basis of texture extracted from image data, that appear to bound an object to be tracked and upon which, therefore, attention must be focused. In laboratory tests, the stereo-cluster approach was demonstrated to enable robust tracking of objects in 3D.
This work was done by Eric Huber and Bryn Wolfe of Metrica, Inc., and Jeff Kerr of JRKerr for Johnson Space Center. For further information, access the Technical Support Package (TSP) free on-line at www.nasatech.com/tsp under the Electronic Components and Systems category.