Promising Directions in Active Vision Eye-hand Coordination Sensors on a Manipulator Can Augment Vision with Integration with Robot Architectures Active Vision Research Should Be Con

iii Summary Active vision systems have mechanisms that can actively control camera parameters such as position, orientation, focus, zoom, aperture and vergence (in a two camera system) in response to the requirements of the task and external stimuli. They may also have features such as spatially variant (foveal) sensors. More broadly, active vision encompasses attention, selective sensing in space, resolution and time, whether it is achieved by modifying physical camera parameters or the way data is processed after leaving the camera. In the active vision paradigm, the basic components of the visual system are visual behaviors tightly integrated with the actions they support; these behaviors may not require elaborate categorical representations of the 3-D world. Because the cost of generating and updating a complete, detailed model of our everyday environment is too high, this approach to vision is vital for achieving robust, real-time perception in the real world. In addition, active control of imaging parameters has been shown to simplify scene interpretation by eliminating the ambiguity present in single images. This document describes promising directions for research in active vision and possible applications of this research. It also discusses progress in experimental equipment for supporting this research and potential applications. Important research areas in active vision include attention, foveal sensing, gaze control, eye-hand coordination, and integration with robot architectures: Attention Selective processing of regions with restricted location, motion, or depth, is necessary for achieving real-time performance with limited resources. Although the design of the attention system is potentially highly complex, and will aaect the design of visual processing at all levels, little is known in detail about what this system should look like. Foveal Sensing A spatially-variant (foveal) sensor permits high resolution at the location of interest without the cost of uniformly high resolution. Issues that need exploring include the properties of unconventional \log-polar" coordinate systems speciically suited to foveal sensing, and the determination of visual features which v can be informative at low resolution to allow reliable selection of targets in the periphery. Gaze Control The alteration of imaging parameters to aid in the performance of visual tasks, or gaze control, is useful for many tasks, including image stabilization, overcoming a limited eld of view, gure-ground separation and range estimation. Gaze control is divided into two primary categories: Gaze stabilization and gaze change. The former consists of controlling the camera to maintain clear images of some world point that …