Distinguishing Pedestrians Facing to the Front and the Side by Gait Observation

In this paper, we propose a method to distinguish pedestrians facing to the front and the side by using a low resolution and quality surveillance image sequence. In the past, there have been many methods to estimate the head orientation of a pedestrian. However, because all these methods use facial texture information to achieve the goal, it is difficult to apply the methods to a low resolution and quality image sequence that does not include enough information. Therefore, we focus on the gait change of a pedestrian, which can be acquired even from a low-resolution silhouette sequence. Experiments confirm the effectiveness of the proposed method by using low-resolution image sequences of over one hundred subjects.