Stereoscopic video generation from monoscopic video

Abstract

With the high development of stereoscopic technology, the stereoscopic video became more and more popular in the daily life since they offered a more realistic sense to viewer than the 2D video. The maturity of the 3D technology enabled a good effect of the stereoscopic sense which lead to the 3D movie obtained a better benefit than the 2D in the movie industry. And now in European, more and more researchers focus on the 3D TV development. In addition, by the end of the 2009, most of the television manufacturers release the 3D-enabled television into the market which enable people enjoy the 3D system at home. All the above facts indicated the 3D dynasty will come sooner than our expectation.

Since a great deal of monoscopic videos stored in many databases, the transformation from the monoscopic video to a stereoscopic image sequence is a problem of considerable practical interest. Converting to the stereoscopic video can offer more realistic sense of the scene to the viewer. In addition, with the innovation of computer technology and image processing, stereoscopic display systems are gradually used in many areas such as medical engineering navigation and entertainment applications. Even if the stereoscopic videos are used in many aspects, the stereo contents are still not rich enough for viewers, especially, for converting the monoscopic video to stereoscopic video. Therefore how to achieve this process becomes the most important topic in stereoscopic research.

Stereoscopic video is relied on the illusion effect of the human eye. Due to the difference of perception of the left and right eyes to reconstructed the stereo effect. The domain stereoscopic video reconstruction is based on the Depth Image Based Rendering (DIBR) which requires an input source video and its corresponding depth map.

The purpose of this project is to generate low quality of stereoscopic video from the monoscopic video. There are two key steps in the stereo video generation, 1) the depth map generation and 2) left and right eye videos generation based on DIBR. The source video is used to estimate the depth map based on motion vectors and segmentation map based on colors. Fusion with depth and segmentation map is a step to get the final depth map. Filtering is applied to final depth map to avoid the clipping effect in the next step, constructing the stereoscopic video based on the DIBR. For user friendly consideration, a graphic user interface is integrated to this project.

Base on the motion parallax, the near objects move faster across the human retina than the distant ones which provides important depth cue. The source video requires background is static and the objects with the movement. Otherwise, it causes significant effect for the depth estimation. Results show that the proposed method achieves good stereo effect for videos with static background. 立体视频最主要是依靠人眼睛的错觉。因为左右两只眼睛的看到不同的景象，从而在人的大脑里产生立体的效果。这个设计的目的是实现把非高清2D视频出转成3D视频。2D转3D主要包括了两个部分：深度图的估算和立体图像的生成（DIBR）。深度图的估算是利用眼睛的运动视差原理去估算物体的深度。原视频首先同时经过运动估计和颜色分割，融合在一起，最后过滤产生出比较理想的深度图。得出的深度图和原视频再经过立体图像重组和填洞，得到最后的3D视频。这个过程叫做DIBR。考虑到用户方便的问题，该设计添加了图形用户界面。 视差原理（近的物体在人的视觉膜上运动得比远的要快）提供了一条很重要的深度信息，利用这个深度信息去估计物体与观看者之间的距离。从而重组左右眼的视频。因为本设计利用了视差原理，所以原视频要求要有运动。结果显示，原视频的背景趋向于静态，那么所得出来的立体效果会比较明显。