Virtual environment stereo image capture using the Unreal Development Kit

Abstract

In this paper we propose a novel method for acquiring stereo images using the Unreal Development Kit. Our method permits the creation and capture of virtual stereo scenes in a carefully controlled environment. Stereo image pairs produced by the system can subsequently be used for testing stereo matching algorithms and a large variety of computer vision techniques. Compared to a physical stereo camera setup this virtual method provides significantly finer and a better quantitative control over all variables involved in image acquisition ranging from camera parameters to specific detail of object materials such as texture, colour and reflectiveness including their behaviour under different lighting conditions. This is illustrated in this paper by comparing images produced via this method to those captured with physical stereo cameras in stereo matching tests using OpenCV algorithms.