Driver behavior classification and warning strategies for stop controlled intersection approaches

Intersections form some of the most complex and dangerous roadway elements, account- ing for about 50% of all major injuries in US road traffic. Many urban intersections in the US are stop controlled and an unaware, inattentive or even deliberate violation of a stop sign may lead to severe accidents. It is argued that a stop violation detection al- gorithm addressing such violations by warning the driver of the hazardous situation can help avoid or mitigate these accidents. The following thesis describes the development and implementation of two new warning strategies, a dynamic warning distance and a support vector machine with zone-based decision. The warning strategies classify the driver’s be- havior during the approach towards a stop controlled intersection and provide a warning to presumably violating drivers. Classification is performed upon the vehicle’s dynamic state and its relative position to the upcoming stop sign, which is provided by an exist- ing stereo vision based stop sign detection. Comparing the strategies with an established warning concept shows that true positive classifications can be increased while false pos- itive warnings decrease at the same time. Most importantly, the warnings can be issued earlier during the approach, providing the driver with more reaction time to be able to stop the vehicle before entering the intersection. The warning strategies are integrated in a research vehicle and a concluding real-traffic evaluation confirms the reliable classi- fication performance and demonstrates the potential for operation in future passenger cars.