文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > 互联网 > 2008-15-A novel method for semi-automatic 2D to 3D video conversion
侵权投诉

2008-15-A novel method for semi-automatic 2D to 3D video conversion

A Novel Method for Semi-automatic 2D to 3D Video Conversion

Chenglei Wu, Guihua Er, Xudong Xie, Tao Li, Xun Cao, Qionghai Dai, IEEE senior member

Broadband Networks & Digital Media Laboratory, Department of Automation, Tsinghua University

wucl07@mails.tsinghua.edu.cn

ABSTRACT

In this paper, we present a novel semi-automatic method for converting monoscopic video to stereoscopic video. An efficient interactive image cutout tool is first used to segment the object-of-interest in the key frames. Then, we assign the initial depth information to the segmented objects. These objects are tracked in the whole video sequence through a bi-directional KLT (Kanade-Lucas-Tomashi) algorithm. In addition, depth interpolation is employed to produce the depth information in the non-key frames. Finally, stereoscopic video is synthesized in consideration of different 3D display types. The experimental results show that our method can fulfill 2D to 3D video conversion both reliably and efficiently.

Index Terms—three-dimensional displays, image segmentation, stereo vision

1. INTRODUCTION

Three-dimensional (3D) video provides a dramatic enhancement in the viewing experience compared with its two-dimensional (2D) counterparts. Due to the huge amount of existing 2D video in many databases, converting 2D video to 3D video is of considerable practical interest not only to academic researchers but also to 3D TV industrialists.

Different approaches have been exploited in the past decades to generate the stereoscopic video. These methods can be generally divided into two categories according to whether depth maps are computed. Depth-based approaches are devoted to estimate the structure of the scene from the original 2D video, which remains a computer vision difficulty [1-6]. The other scheme is to synthesize the binocular parallax in stereoscopic video directly from monocular video without computing the scene depth information [7-9]. This kind of depth-free methods allow to convert 2D images into binocular 3D images by selecting images which could be a stereo pair of image among the original sequential images [7]. This idea is improved to get the suited right image to the left of stereo frame pair by

This work was supported in part by The Distinguished Young Scholars of NSFC, No.60525111 and in part by the 863 Program, 2007AA01Z332

performing a planar transformation [8]. [9] formulates the stereo synthesis as an optimization problem via introducing a cost function, which is used to select the most suitable frames in the input video for generating the stereoscopic video frames. However, these methods require the captured scene or objects remain almost stationary and the motion of camera must introduce a horizontal parallax, which restricts the practical implementation to certain scenario.

2D to 3D video conversion by generating a depth map from the original 2D image sequence has been widely studied in an active computer vision area called Structure-From-Motion (SFM). Correspondences between subsequent frames (or similar views of the same scene) are used to recover camera parameters and determine the depth [2]. Using the extended Kalman filter, camera motion is estimated in rotation and translation as well as the structure of scene is estimated [3][4]. For this approach, it is also required that the 3D scene must be predominately static. Another scheme is to calculate the depth maps through analyzing the motion of objects, which assumes that the motion of all objects is proportional to their distance from the camera [5][6]. This assumption only holds in a relatively small proportion of footage encountered. The state-of-the-art methods are mainly automatic 2D to 3D video conversion, while the intrinsic illness of the conversion problem prevents these methods from producing satisfactory results for arbitrary situation.

Since the depth map can be highly compressed and its resolution may be lower than the associated 2D image, these characteristics prompt us to implement the 2D to 3D conversion though depth-based scheme. The bottleneck in the 2D to 3D video conversion lies in the estimation of depth maps from the monoscopic sequence, a mathematically ill-posed problem which admits an infinite number of solutions since true depth information of the scene is unknown. As a result, we integrate user’s interactive operation into our conversion method to gain more information about the scene structure. In this paper, a novel framework for interactive 2D to 3D video conversion is proposed. Key frames are selected for user’s interactive object selection and depth assignment. To obtain the objects’ depth information in non-key frames, a modified algorithm for object tracking is employed, which utilizes the bi-directional tracking and active contour model called snake [10] to track the objects’ contour throughout the video

978-1-4244-1755-1/08/$25.00 ©2008 IEEE

3DTV-CON'08, May 28-30, 2008, Istanbul, Turkey

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email