Driver behavior modeling

Project results: Parametrzation of a driver behavior model

• Analysis of 18 data sets with 16 sections each in less than one day
• Parametrization of a driver behavior model based on the analysis results
• Support for implementation of the driver model in a test car

Driver behavior modeling plays an important part during the development of driver assistance systems and a driver-friendly transition to automatic driving. For example, the prediction of intended driving maneuvers and reaction times allows the software developers to adjust the timing of assistance systems and to initiate or cancel interventions.

As part of a the UR:BAN project with Fraunhofer IAO and BMW Group, we have analyzed the eye movement behavior, when drivers pass pedestrians at the sidewalk. In average 15 % of all road fatalities are attributed to pedestrians each year. Euro NCAP rules are including active pedestrian protection systems. This is the reason why the understanding of pedestrian perception and driver behavior modeling is an important research question for road safety. Figure 1 (left) shows two typical traffic situations with a pedestrian, on the right side a warning cascade is presented to avoid a collision between the car and the pedestrian.

Figure 1: The understanding of visual perception of pedestrians is important to adjust the timing of assistance systems and to initiate or cancel interventions.

The key technology for analyzing the visual perception of pedestrians is eye tracking. In our project Blickshift was responsible for analyzing eye movements from an eye tracking experiment. The goal was to define parameters of a state transition model, which decides to warn a driver about a pedestrian, see Figure 2.

Figure 2: By analyzing the eye movement behavior, a state transition model has been parametrized in order to warn a driver about a pedestrian.

In a driver simulation experiment 18 participants have driven nine minutes the same route. The route was divided into 14 sections, representing different traffic situations with and without pedestrians. In some sections the pedestrians did not moved, in others they have moved towards the road with different velocities. Besides eye movements, pedal activity was recorded.

The critical parameter of the model was the distance between the fixation position in the traffic and the position of the pedestrian. The hypothesis was, that the more often this distance was below a certain threshold, the higher the probability was, that the driver has perceived the pedestrian. Our job was to make a proposal about this threshold.

Figure 3: The model parameters have been defined by using a concatenation of Value Search components in Blickshift Analytics together with a visualization of the recorded data with Line Graphs and Bar Charts.

Diederichs, Frederik ; Brouwer, Nina ; Klöden, Horst ; Zahn, Peter ; Schmitz, Bernhard: Application of a Driver Intention Recognition Algorithm on a Pedestrian Intention Recognition and Collision Avoidance System. In: Bengler, K. ; Drüke, J. ; Hoffman, S. ; Manstetten, D. ; Neukum, A. ; Bengler, K. ; Drüke, J. ; Hoffman, S. ; Manstetten, D. ; Neukum, A. (Hrsg.) ; Bengler, K. ; Drüke, J. ; Hoffman, S. ; Manstetten, D. ; Neukum, A. (Hrsg.): UR:BAN Human Factors in Traffic – Approaches for Safe, Efficient and Stress-free Urban Traffic, UR:BAN Human Factors in Traffic – Approaches for Safe, Efficient and Stress-free Urban Traffic : Springer Fachmedien, 2017

Raschke, Michael ; Schmitz, Bernhard ; Wörner, Michael ; Ertl, Thomas ; Diederichs, Frederik: Application Design for an Eye Tracking Analysis based on Visual Analytics. In: Proceedings of the Symposium on Eye Tracking Research and Applications, ETRA ’16, Proceedings of the Symposium on Eye Tracking Research and Applications, ETRA ’16, 2016.