基于运动估计的2D转3D视频深度滤波  

2D to 3D Video Depth Filter Based on Motion Estimation

作　　者：黄冬冬[1] 张贻雄[1] 石江宏[1] 

机构地区：[1]厦门大学信息科学与技术学院,福建厦门361005

出　　处：《厦门大学学报（自然科学版）》2013年第4期473-478,共6页Journal of Xiamen University：Natural Science

基　　金：国家自然科学基金项目(61102135);中央高校基本科研业务费(2010121063);中国博士后科学基金项目(2012M511433)

摘　　要：根据视频中距离越近运动尺度越大的原理,物体的运动包含了提取2D视频深度的最有效信息.然而自然视频中,物体运动存在加速度,物体在同样深度情况下,运动时大时小,结果将导致通过运动估计提取的深度存在不连续性.根据物体运动连续性的特点,同一物体在不同帧中的深度变化也存在连续性.提出一种基于运动估计的2D转3D深度滤波算法.该算法中,以高斯滤波为基础,并采用相似度作为权重,自适应地修改高斯滤波器的参数,实现了自适应的高斯滤波.实验结果表明,采用该算法有效地减小了运动估计误差,平滑了深度序列,提高了深度序列的准确性和合理性.Depth information extracting is one of the important step in 2D to 3D video transform. Accurate depth information make it easier in 3D video transform later. According to the principle of the closer object has larger movement dimension in video,motion of object contains the most effective information in getting depth of 2D vedio. However, because object in original video has acceleration, the speed is different on the same depth. It results discrete depth when calculating the depth information through movement informa- tion. According to the characteristic of objectrs continuous movement,object has successive depth in different frames. In this papaer, we propose a depth filter method based on motion estimation for 2D to 3D transforming. This algorithm based on gauss filter and ada- ptationally change the parameter of gauss filter according to similarity. Finally,we achieve a auto-adapted gauss filter. Experimental results show that the proposed method reduce the error of motion estimationan and smooth the depth serials,improve the accuracy and rationality of depth serials.

关 键 词：深度提取 关键点滤波 运动估计 运动矢量 深度滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]