Vision in and out of Vehicles

12 1094-7167/03/$17.00 © 2003 IEEE IEEE INTELLIGENT SYSTEMS Her baby daughter was delivered seven weeks early by emergency Caesarian section and is in critical condition. The girl’s 10-month-old brother, who was in a child restraint in the backseat, is also in critical condition. The accident happened when a car, thought to have been traveling at up to 80 kmph (the speed limit), failed to stop. Days later, another fatality occurred—this time a truck driver. The truck left its lane, jumped the guard rail, and hit an embankment. Both accidents happened in fine driving conditions on the city’s edge. Perhaps the drivers could have avoided these accidents if they had received a warning about the impending situation. Almost every driver has experienced a warning from a passenger, perhaps about an obscured car while merging or a jaywalking pedestrian in a blind spot. Such warnings could save countless lives every day. We believe that, in the near future, such assistance will come from the vehicle itself. Many vehicles already employ computer-based driver assistance in the form of antilock braking systems or adaptive engine management systems. However, more than a decade after autonomous-system technologies emerged, systems such as those that Universität der Bundeswehr Munich1 and Carnegie Mellon University’s NavLab group2 developed have not been realized commercially. A key distinction between existing systems and promising R&D systems is driver and vehicle manufacturer acceptance. Because a paradigm shift to autonomous vehicles is unlikely, intelligent-vehicle technologies must make it to the road by subsystems instead. These subsystems should solve a small, well-defined task that supports, not replaces, the driver. At the Australian National University’s Intelligent Vehicle Project, we are developing such subsystems for