Page 36 - Dynamic Vision for Perception and Control of Motion

20 1 Introduction

Based on experience gained in these areas, Chapter 12 discusses sensor re-
quirements for advanced vision systems in automotive applications and shows an
early result of saccadic perception of a traffic sign while passing. Chapters 13 and
14 give an outlook on the concept of such an expectation-based, multifocal, sac-
cadic (EMS) vision system and discuss some experimental results. Chapter 13 pre-
sents the concept for a dynamic knowledge representation (DKR) serving as an iso-
lation layer between the lower levels of the system, working mainly with methods
from systems dynamics/engineering, and higher ones leaning mainly on “artificial
intelligence” methods. The DOB as one part of DKR is the main memory for all
objects and subjects detected and tracked in the environment. Recent time histories
of state variables may be stored as well; they alleviate selecting the most relevant
objects/subjects to be observed more closely for safe mission performance. Chapter
14 deals with a few aspects of “real-world” situation assessment and behavior–
decisions based on these data. Some experimental results with this system are
given: Mode transition from unrestricted roadrunning to convoy driving, multi–
sensor adaptive cruise control by radar and vision, autonomous visual lane
changes, and turnoffs onto crossroads as well as onto grass-covered surfaces; de-
tecting and avoiding negative obstacles such as ditches is one task solved in cross-
country driving in a joint project with U.S. partners.
Chapter 15 gives some conclusions on the overall approach and an outlook on
chances for future developments.