北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2010, Vol. 33 ›› Issue (4): 26-29.doi: 10.13190/jbupt.201004.26.yuanh

• 论文 • 上一篇    下一篇

保持空间准确度与时间一致性的深度估计算法

元辉,常义林,刘晓仙,刘河潮,卢朝阳   

  1. 西安电子科技大学
  • 收稿日期:2009-09-25 修回日期:2010-02-04 出版日期:2010-08-28 发布日期:2010-05-21
  • 通讯作者: 元辉 E-mail:yuanhui0325@gmail.com,hyuan@mail.xidian.edu.cn
  • 基金资助:

    国家自然科学基金(60772134)、高等学校学科创新引智计划(B08038)、陕西省自然科学基金(SJ08F03)和西电-华为多媒体通信联合实验室合作专项基金资助课题;国家级.国家自然科学基金项目

A Depth Estimation Algorithm for Preserving Spatial Accuracy and Temporal Consistency

  • Received:2009-09-25 Revised:2010-02-04 Online:2010-08-28 Published:2010-05-21

摘要:

为提高深度图的准确度,提出了一种基于图割的深度估计算法. 利用平均亮度梯度联合匹配测度函数代替单像素亮度匹配,并考虑邻近像素深度相关性改变图割网络结构,提高深度估计的空间准确度;采用前一时刻同位置像素的深度约束当前时刻像素的深度值,使估计的深度图在时间上保持一致. 提出的算法考虑了深度图的空间、时间特征,可以提高深度图的准确度. 实验结果表明,与传统的基于图割的深度估计算法相比,新算法估计的深度图更加准确.

关键词: 深度估计, 图割, 空间准确度, 时间一致性

Abstract:

A new depth estimation algorithm based on graph cuts is proposed to improve the accuracy of depth maps. Average luminancegradient matching method is proposed to replace unitary pixel luminance matching. The graph structure is changed considering depth correlations between each adjacent pixel. The depth of the colocated pixel in previous frame is used to restrict the depth of current pixel. Taking consideration of the spatial and temporal features of depth maps, the accuracy of estimated depth maps is then improved. Experimentss demonstrate that the proposed method can generate more accurate depth maps than the traditional graph cuts in algorithms.

Key words: depth estimation, graph cuts, spatial accuracy, temporal consistency

中图分类号: