Movement direction-based approaches for pedestrian detection in road scenes

Pedestrian Detection is a critical technique for avoiding the collision between the vehicle and people, and it can be used in the advanced driver assistance system. Most research of the pedestrian detection areas are focused on the standing or walking people at the training process. INRIA´s pedestrian dataset is composed of persons standing and facing the front, however another datasets comprise various types of pedestrian without classification for direction. In other words, movement directions of the pedestrian are not considered on creating detectors. In this paper, we propose a pedestrian detection method using pedestrian data classified into four by moving directions such as front, back, left and right. Each of detectors created by categorized data are integrated, which are used for pedestrian detection. For the training, we use histograms of oriented gradients using the direction distribution of the edges. In the experiments, we use the pedestrian datasets obtained by moving vehicle in order to enhance public confidence. Our result shows the improved detection ratio in comparison to existing methods underutilized the moving direction.