Detecting unusual pedestrian behavior toward own vehicle for vehicle-to-pedestrian collision avoidance
Kota Nakatsubo, Keiichi Yamada · 2010
Pedestrian protection is an important issue for intelligent vehicles. This paper proposes a new approach for predicting the possibility of a collision between a vehicle and a pedestrian. Almost all pedestrian behavior toward vehicles observed in the real world is considered safe. Therefore, pedestrian behavior that deviates from usual pedestrian behavior indicates a possibility where the vehicle must take urgent evasive action to avoid collision with the pedestrian. From such a viewpoint, this paper proposes a method for predicting whether the pedestrian behavior deviates from usual pedestrian behavior. Usual pedestrian behavior as observed from the vehicle is modeled with machine learning to detect whether a new observed behavior deviates from the model of the usual pedestrian behavior. The effectiveness of the proposed method is demonstrated with an experiment conducted in a simple road environment.