Vision-based system identification and state estimation

Abstract

There are many situations where a (primary) controlled system must “interact” with another (secondary) system over which it has no direct control; e.g. the robot arm of a space vehicle grasping some free-floating object, a plane being refueled in flight by a tanker aircraft, a military tank intent on identifying and destroying enemy tanks, a surgeon attempting to remove all portions of an oddshaped tumor, and a highway vehicle trying to maintain a certain speed as well as a safe distance from other vehicles. Such scenarios can involve both stationary and moving objects. In virtually all cases, however, the more knowledge that the primary system has about the secondary system, the more probable the success of the interaction. Clearly, such knowledge is very often vision-based. This paper will focus on some recent results related to both identifying what a planar object (system) is and what its static or dynamic state is, based primarily on different views of its boundary. Boundary data information has been used extensively in a wide variety of situations in pattern analysis and image understanding. While the results that we will present here also are more generally applicable in computer vision, we will focus on how they can be applied to control system applications, and more specifically to the “visual” part of “visual-servoing.”