System-Theoretic Aspects of Dynamic Vision

Abstract

Our aim in this minicourse is twofold. On one hand, we aim to show that the introduction of computer vision in control systems, i.e. “Vision in the Loop”, raises exciting and yet unexplored problems in system theory. On the other hand, we explain how tools from control and estimation theory are, nowadays applied [3, 6, 7, 26, 27, 38, 54, 57, 63, 66] to “Dynamic Vision” problems with rather encouraging results in traditionally difficult applications, such as autonomous vehicle navigation [17, 18, 19], vision-based tracking and servo [9, 35, 36, 50], vision-based manipulation [5, 35, 36], docking [18, 42], vision-based planning [12], active sensing [67]. The course is therefore divided in two parts. In the first part, we shall pose two fundamental dynamic vision problems, namely 3-D motion and scene structure recovery, in a system theoretical framework. In the second part, we shall illustrate some relevant applications of dynamic vision to control systems.