Page 5 - Dynamic Vision for Perception and Control of Motion

Preface

During and after World War II, the principle of feedback control became well un-
derstood in biological systems and was applied in many technical disciplines to re-
lieve humans from boring workloads in systems control. N. Wiener considered it
universally applicable as a basis for building intelligent systems and called the new
discipline “Cybernetics” (the science of systems control) [Wiener 1948]. Following
many early successes, these arguments soon were oversold by enthusiastic follow-
ers; at that time, many people realized that high-level decision–making could
hardly be achieved only on this basis. As a consequence, with the advent of suffi-
cient digital computing power, computer scientists turned to quasi-steady descrip-
tions of abstract knowledge and created the field of “Artificial Intelligence” (AI)
[McCarthy 1955; Selfridge 1959; Miller et al. 1960; Newell, Simon 1963; Fikes, Nilsson
1971]. With respect to achievements promised and what could be realized, a similar
situation developed in the last quarter of the 20th century.
In the context of AI also, the problem of computer vision has been tackled (see,
e.g., [Selfridge, Neisser 1960; Rosenfeld, Kak 1976; Marr 1982]. The main paradigm ini-
tially was to recover a 3-D object shape and orientation from single images (snap-
shots) or from a few viewpoints. On the contrary, in aerial or satellite remote sens-
ing, another application of image evaluation, the task was to classify areas on the
ground and to detect special objects. For these purposes, snapshot images, taken
under carefully controlled conditions, sufficed. “Computer vision” was a proper
name for these activities since humans took care of accommodating all side con-
straints to be observed by the vehicle carrying the cameras.
When technical vision was first applied to vehicle guidance [Nilsson 1969], sepa-
rate viewing and motion phases with static image evaluation (lasting for minutes
on remote stationary computers in the laboratory) had been adopted initially. Even
stereo effects with a single camera moving laterally on the vehicle between two
shots from the same vehicle position were investigated [Moravec 1983]. In the early
1980s, digital microprocessors became sufficiently small and powerful, so that on-
board image evaluation in near real time became possible. DARPA started its pro-
gram “On strategic computing” in which vision architectures and image sequence
interpretation for ground vehicle guidance were to be developed (‘Autonomous
Land Vehicle’ ALV) [Roland, Shiman 2002]. These activities were also subsumed
under the title “computer vision”, and this term became generally accepted for a
broad spectrum of applications. This makes sense, as long as dynamic aspects do
not play an important role in sensor signal interpretation.
For autonomous vehicles moving under unconstrained natural conditions at
higher speeds on nonflat ground or in turbulent air, it is no longer the computer
which “sees” on its own. The entire body motion due to control actuation and to