Page 6 - Dynamic Vision for Perception and Control of Motion

vi Preface

perturbations from the environment has to be analyzed based on information com-
ing from many different types of sensors. Fast reactions to perturbations have to be
derived from inertial measurements of accelerations and the onset of rotational
rates, since vision has a rather long delay time (a few tenths of a second) until the
enormous amounts of data in the image stream have been digested and interpreted
sufficiently well. This is a well-proven concept in biological systems also operating
under similar conditions, such as the vestibular apparatus of vertebrates with many
cross-connections to ocular control.
This object-oriented sensor fusion task, quite naturally, introduces the notion of
an extended presence since data from different times (and from different sensors)
have to be interpreted in conjunction, taking additional delay times for control ap-
plication into account. Under these conditions, it does no longer make sense to talk
about “computer vision”. It is the overall vehicle with an integrated sensor and
control system, which achieves a new level of performance and becomes able “to
see”, also during dynamic maneuvering. The computer is the hardware substrate
used for data and knowledge processing.
In this book, an introduction is given to an integrated approach to dynamic vis-
ual perception in which all these aspects are taken into account right from the be-
ginning. It is based on two decades of experience of the author and his team at
UniBw Munich with several autonomous vehicles on the ground (both indoors and
especially outdoors) and in the air. The book deviates from usual texts on computer
vision in that an integration of methods from “control engineering/systems dynam-
ics” and “artificial intelligence” is given. Outstanding real-world performance has
been demonstrated over two decades. Some samples may be found in the accom-
panying DVD. Publications on the methods developed have been distributed over
many contributions to conferences and journals as well as in Ph.D. dissertations
(marked “Diss.” in the references). This book is the first survey touching all as-
pects in sufficient detail for understanding the reasons for successes achieved with
real-world systems.
With gratitude, I acknowledge the contributions of the Ph.D. students S. Baten,
R. Behringer, C. Brüdigam, S. Fürst, R. Gregor, C. Hock, U. Hofmann, W. Kinzel,
M. Lützeler, M. Maurer, H.-G. Meissner, N. Mueller, B. Mysliwetz, M. Pellkofer,
A. Rieder, J. Schick, K.-H. Siedersberger, J. Schiehlen, M. Schmid, F. Thomanek,
V. von Holt, S. Werner, H.-J. Wünsche, and A. Zapp as well as those of my col-
league V. Graefe and his Ph.D. students. When there were no fitting multi-
microprocessor systems on the market in the 1980s, they realized the window-
oriented concept developed for dynamic vision, and together we have been able to
compete with “Strategic Computing”. I thank my son Dirk for generalizing and
porting the solution for efficient edge feature extraction in “Occam” to “Transput-
ers” in the 1990s, and for his essential contributions to the general framework of
the third-generation system EMS vision. The general support of our work in “con-
trol theory and application” by K.-D. Otto over three decades is appreciated as well
as the infrastructure provided at the institute ISF by Madeleine Gabler.

Ernst D. Dickmanns