Eulerian-Lagrangian stereo-to-multi-view Conversion

Abstract

In contrast to the popularity of stereoscopic 3D (S3D) movies in movie theaters, the adoption of S3D at home is low. It is widely believed that watching S3D with glasses is not the pratical approach for a home setting. A much more appealing approach is to use automultiscopic displays that provide a glasses-free 3D experience to multiple viewers. The main technical challenge that hampers the adaptation of this technology is the lack of multi-view content. We develop a real-time system that converts stereoscopic video to a high-quality, multi-view video, which can be directly fed to automultiscopic displays. Our algorithm uses a wavelet-based decomposition of stereoscopic images with per-wavelet disparity estimation. One key contribution lies in combining Lagrangian and Eulerian approaches. This allows us to leverage their complementary advantages. The method achieves real-time performance on current GPUs. Its design also enables an easy hardware implementation, demonstrated using a field-programmable gate array.