An Integrated Adas Solution for Pedestrian Collision Avoidance

Accident statistics indicate that pedestrians constitute a large share of vehicle-related fatalities worldwide. Due to continuing trends towards urbanization, this proportion can be expected to further increase. Advances in passive safety have already proven their effectiveness, but since injuries cannot be completely avoided at higher collision speeds a preferred solution is the complete avoidance of collisions. In this paper, we introduce an active safety approach for preventing collisions with pedestrians that integrates advanced perception systems and executes emergency braking and steering maneuvers. The functional concept and system architecture are introduced, followed by the design of the actuation setup. Finally, the results of extensive driving tests are given for validation. As part of the validation strategy, a testing facility has been constructed that comprises a horizontal truss with a pedestrian dummy suspended beneath it. This pedestrian dummy can be moved laterally to simulate pedestrian motion. The presented system architecture includes abstract levels for sensorics, perception refinement, situation analysis and actuation. The functionality is realized using a stereo camera and radar, both of which are high-performance, stateof-the-art automotive sensors currently in series production. The stereo camera integrates a pedestrian classification algorithm, and together the sensors provide extensive knowledge about the available maneuvering space. The sensor data are combined into a hybrid environment representation with two separate entities for moving objects and static structures. This representation can be used as a basis for the situation analysis logic, determining if an emergency braking or steering maneuver is necessary. Two actuators are used to facilitate maneuver execution: an electric power steering (EPS) system and an innovative brake system specifically designed for a fast and precise electronic actuation. One algorithm implemented for handling pedestrian scenarios is the pedestrian motion prediction. In these cases, commonly-used models for vehicle motion are no longer valid, so a motion prediction algorithm has been developed that specifically considers pedestrian behavior. The result, as demonstrated in relevant scenarios, is a significant decrease in false-positive system reactions. In this paper, possibilities for how an emergency situation can evolve with respect to available maneuvering space and last point to brake or steer are extensively discussed and examined through driving tests. An additional challenge is the handling of scenarios where a pedestrian assumes a more generic appearance, such as a person using a wheelchair or pushing a stroller. A holistic system for avoiding pedestrian accidents has been designed, implemented and extensively tested. The results quantitatively show the benefits in terms of the detection performance of the environmental sensors and the sophisticated environment model, including information about the available maneuvering space. Classification and prediction algorithms have been implemented that take into account the characteristics of pedestrian behavior to determine the desired system reactions. Since all sensors and actuators are currently in or near series production, the presented approach demonstrates how pedestrian safety can be greatly enhanced in the near future.