Abstract

This paper tackles one of the first causes of death all around the world. In fact, third world countries including Morocco suffers more from road accidents caused by undisciplined human behavior specially pedestrians. To address this issue, our work as part of a road safety project named SAFEROAD, aims to detect pedestrians and classify their walking direction in order to alert the driver of their presence using a new neural network algorithm named Capsule Network.
As we work for the Moroccan case, we collected the first Moroccan pedestrian dataset from several Moroccan cities so as to consider the Moroccan pedestrian behavior in the proposed system. The collected dataset was used to train the proposed network giving an accuracy of 78.95% on the validation phase. The proposed approach shown better results compared to convolutional neural network algorithms.