I am interested in building working vision systems, both in robots and for using visual information for science, medicine, and other applications. I am especially interested in early vision, especially computation of scene structure through stereo and motion, and the integration of vision modules to produce robust systems.

I am studying sensing and action for robots in dynamic environments, under a variety of operating constraints. The problems to be solved are not just system configuration; there are many open issues in determining the proper interactions between the vision component of a robot and the reasoning and action components. This requires investigating how world modeling requirements and computing system facilities can be combined into a vision-based operating system. These concepts are embodied in tracking systems, navigating mobile robots and other robotic systems.

Two recent strong interests are seeing visual motion and adaptive triangulation for terrain representation. In the first, Jeff Boyd and I have built a system to recognize walking people using periodic variation in the 'shape of motion', a description of the instantaneous distribution of motion in a moving figure. In the second, I use curvature descriptions of surfaces to isolate terrain features, which form the skeleton of a triangulated mesh approximating the terrain. This requires work in multiscale descriptions of surfaces, surface fitting methods, and applications of computational geometry.