Method for evaluating human and simulated drivers in real traffic situations

Understanding the characteristics of human driving behavior as well as the situational and emotional factors affecting it is important for the design of information, warning and driving enhancement systems. In-car control data as well as data representing the surrounding context are crucial to establish such an understanding. Our approach focuses on the latter by presenting a general method to extract vehicle trajectories from video taped intersections or roadways. It uses the Kolmogorov-Smirnov test on time-based and space-based cumulative distribution functions of a given pixel to differentiate between background and moving vehicles. Simulated cars following the extracted trajectories (i.e. position, velocity and heading) together with the scene layout are embedded into our driving simulator. By assigning human (or model) control to one of the cars, comparisons can be made between the behavior of the tracked car and that of different subjects (or driving models) as they are exposed to the same realistic traffic situation and behavior