Video Segmentation For2D To3D Video Conversion

Video segmentation is an important problem in Computer Vision, and has various applications in file making, such as film post production. With the development of multi-media technology,3D video appears to be more and more popular, and academia have focused on converting monocular videos to3D videos. In recent years, many researchers have proposed methods using machine learning technology for video conversion, which are highly automatic. However, the quality of the converted3d videos is not good enough, and most of these methods are limited to some scenes, such as streetscape or indoor scene. Methods based on video segmentation can guarantee the quality of converted videos, although need more user interaction.The primary problem of the2D to3D technology for monocular video based on video segmentation is how to obtain good video segmentation results effectively, how to achieve consistency in space and time domain, and how to deal with semitransparent objects and slender objects, such as hair. In this paper, a semi-automatic segmentation tool based on user interaction is implemented. The tool uses color, shape and motion information of objects to obtain good segment results, and with the help of redundant information in videos, it makes the process much more effective. We also proposed a novel segmentation method for static objects, which further improves the efficiency of the system. In order to meet the requirements of2D to3D video conversion, we developed a new method to make segmentation more temporally consist. The experiments show that the proposed system can yield good segmentation and is practical and robust.