Dickmanns 2003: Based on more than a decade of experience with autonomous
road vehicles, a third-generation dynamic vision system has been developed over
the last three years. The ‘MarVEye’ camera configuration realized with three to
four conventional miniature CCD-TV-cameras mounted fix relative to each other
on a high bandwidth pan-and-tilt platform allows new levels of performance. A
saccadic type of vision with intermittent phases of fast saccades and smooth
pursuit does complicate image sequence interpretation; however, with explicit
spatio-temporal representations available in the 4-D approach this can be
handled in a straightforward way. A coherent form of representation of dynamic
scenes containing multiple independently moving objects has been achieved
through homogeneous coordinate transformations as edges in a scene tree.
The system based on COTS-PC-hard- and software, visual / inertial data fusion and object-oriented programming is presented. Applications discussed are: I. Turning-off onto a crossroad in a network of minor roads without lane markings, and 2. hybrid adaptive cruise control on high-speed roads with radar / vision data fusion.