GPU 기반의 실시간 포비티드 렌더링

Abstract

Today's Head-Mounted Displays and high-resolution display above 4K provide a high visual experience and an immersive experience for users. However, the rendering on high resolution displays needs efficient rendering techniques enough to give a real-time. One of such techniques is a foveated rendering combined with gaze-tracking that minimizes the loss of perceptual detail. Therefore, we propose a GPU based real-time foveated rendering technique. The proposed method operates in a way that imitates the human visual system. We designed the sampling map considering the human retina, visual attention, and efficiency of reconstruction. Especially, it can be designed to effectively work on the GPU, and also reduces the computational cost that has been improved by previous foveated sampling. In addition, we use a combination of ray tracing and temporal reprojection to generate a complemented sparse pixel image. Finally, we generate a dense image through the reconstruction process. We also show real-time foveated rendering pipelines for various purpose that are effectively mapped on the GPU. As a result, the proposed method allows high-quality generated images in real time on a high-resolution display with minimizing loss of perceptual detail.