The ability to drive under computer control in a variety of environments
is critical to all mission scenarios envisioned for unmanned ground vehicles.
However, navigation in outdoor terrain is difficult due to lack of easily
and uniquely identifiable landmarks. This paper describes a field-capable
system for navigation, obstacle avoidance, simulated visual training of
mobile robots, world perception modeling using visual feedback information
for the purpose of terrain learning. In this work, a method for lane marker
tracking is described. The method addresses the needs of robustness and
real-time performance. In the area of reliability, the method is strongly
strengthened by performance prediction using the FMCell software simulation.
This method has been used for autonomous driving of the simulated TSU TIGER.
The algorithm performs well in the presence of continuous, dashed, alternating
dashed and missing lane markers. In the approach presented here, the road
is represented by a 2D model in the image plane with camera calibration
and coordinate transformations. Results of simulation runs illustrating
the capabilities of this technique are provided. The technique provides
a better and simplified approach visual servoing of UGV’s.
