动态规划立体匹配的状态空间分析和性能改进  

State space analysis and performance improvement for stereo matching based on dynamic programming

作　　者：杨致中[1] 西宝[1] 

机构地区：[1]哈尔滨工业大学管理学院,黑龙江哈尔滨150001

出　　处：《应用科技》2015年第5期19-23,共5页Applied Science and Technology

基　　金：高等学校博士学科点专项科研基金资助项目(20121108110002)

摘　　要：作为立体视觉的核心技术,立体匹配成为机器视觉领域的热点研究。为提高动态规划立体匹配方法的性能,构建了状态空间模型,分析视差搜索范围、聚合窗口大小、控制点个数这3个主要影响因素。改进方案由2个部分构成:控制点的求取放在高斯塔图像的低分辨率层级上进行,聚合窗口的形状根据两侧控制点的位置进行调整。实验结果表明,经过改进后的动态规划立体匹配方法,可以在确保视差图像质量的情况下,更加快速地完成立体匹配过程。As a key technology for stereo vision, stereo matching has become a hot research topic in the field of ma-chine vision. In order to improve the performance of the stereo matching method based on dynamic programming, astate space model was built to analyze three main influencing factors, including disparity search rangeclustering window, and the number of control points. The improved scheme is composed of two parts：points are computed at a low resolution level image of the Gauss tower, and the shape ofcording to the position of control points at two sides. The experimental results show thatquickly finish the stereo matching process while guaranteeing the quality of thedisparity

关 键 词：动态规划 立体匹配 状态空间分析 控制点 聚合窗口 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]