CNN-Based Pedestrian Orientation Estimation from a Single Image

Abstract

In traffic environments where both vehicles and pedestrians coexist, predicting the path of a pedestrian is an important task for automated driving and driver support systems to prevent accidents. Therefore, research has been conducted to estimate the orientation of a pedestrian using in-vehicle camera images. In this paper, we present a CNN-based method of estimating the pedestrian orientation from single-frame images. The proposed method focuses on the fact that there is a relationship between the direction of a pedestrian's body and the direction of the pedestrian's face. The method is evaluated using TUD and PDC datasets, and the performance is shown.