Real-time Reconstruction of PseudoWide-angle Images with an Approximating Depth Model

Abstract

Recently, interactive contents with high quality images are widely used in various areas. One of the ways that makes the contents more realistic is displaying them as covering viewer's field-of-view. However, the view-angle of the existent contents are not enough for covering the large display surfaces because they are based on household TV monitors whose target view-angle is about 30 degrees. In order to achieve realistic feeling of being in that content, we have to enlarge the view-angle of these images in real-time. So, in this research, we propose a technique that reconstructs wide view-angle images by extracting peripheral images from time-series image frames in realtime. A simple 3-D model approximating target scenes contributes to, achieving real-time processing and reduction of uncomfortable feeling with distortions of the reconstructed peripheral images. We apply this method to an actual game content, and consider the quality of image and its processing time. By using that implementation, we perform objective and subjective evaluations, and confirm the effectiveness for enhancing the fun of the content itself.