Autonomous evasive maneuvers triggered by infrastructure-based detection of pedestrian intentions

We present an active pedestrian protection system that performs an autonomous lane-keeping evasive maneuver in urban traffic scenarios when collision avoidance by braking is no longer possible. The system focuses on pedestrians standing at the curb and intending to cross the street despite an approaching car. It is demonstrated that the evasive maneuver of the car can be initiated before the pedestrian's foot hits the lane, by means of video-based motion contour histograms of oriented gradients and stationary detection. Using clothoid-based real-time trajectory planning and a lateral control of the car, combining feedforward and feedback control, the difference between the driven and the calculated trajectories is kept below 10 cm at maximum lateral accelerations of 4 ms-2 and -5 ms-2. We present the technical realization of the system and its precision with respect to intention recognition and driven trajectories. A case study showed that the system reacted faster than human drivers in five out of 11 cases, with an average time gain of 214 ms, even though the drivers were able to pay the utmost attention to the behavior of the crossing pedestrian.