Modelling Pedestrians Using Artificial Neural Networks

There has been an increasing interest in pedestrian detection in the last decade to save the numerous deaths caused by accidents in which pedestrians are involved. Vehicle manufacturers are addressing these challenges by looking into extendable vehicle body structures , to be activated on first impact with a pedestrian. A complementary approach is to focus on sensor based solutions, which enable vehicles to ”look ahead” and detect pedestrians in their surroundings. Sensor based approaches require a model of pedestrians to validate the measurement against. Modelling pedestrians is especially hard because of the wide range of possible pedestrian appearances. In this paper we have reported some preliminary experiments to demonstrate the feasibility and strength of artificial neural nets to learn and detect pedestrians. Our neural net based model achieves up to 97% classification accuracy.