基于特征点能量的双目立体匹配稳健算法  

Binocular Stereo Matching Robust Algorithm Based on Energy of Marked Points

作　　者：姚海滨[1] 翁金平 崔海华[2] 程筱胜[2] 

机构地区：[1]扬州市职业大学机械工程学院,江苏扬州225012 [2]南京航空航天大学机电学院,江苏南京210016

出　　处：《激光与光电子学进展》2016年第2期134-139,共6页Laser & Optoelectronics Progress

基　　金：国家自然科学基金(51305201;51205192);国家863计划(SS2013AA040801-02);江苏省自然科学基金(BK20130793);江苏省博士后科学基金(1301104C)

摘　　要：为提高双目立体视觉测量图像精确匹配的稳定性,提出一种基于特征点能量的稳健匹配新算法。该算法基于极线约束获取左右图像中特征点的初始匹配,根据特征点之间极线约束关系定义了一种不受仿射变换影响的能量来描述特征点,通过比对特征点能量值来剔除误匹配。该算法有效降低了误匹配率和误剔除率,从而满足多视角测量数据拼合过程中严格限制误匹配率的要求。实验结果表明,该算法正确匹配率大于95%,误剔除率小于2%,具有较强的稳健性。In order to improve the stability of the binocular stereo vision measuring accuracy image matching. A new robust matching algorithm based on feature point energy is proposed, the algorithm based on epipolar constraint to get the initial matching feature points around image. The energy not affected by affine transformation is defined to describe feature points based on epipolar constraint relationship among feature points, and the error matching is removed by comparing the energy of feature points. The proposed method effectively reduces the false matching rate and the false rejected rate so that it can meet the need of strictly limited false matching rate in the process of multiple angle measurement data. Experimental results show that the correct matching rate of the proposed algorithm is over 95%, and the false rejected rate is less than 2%.The algorithm has better robustness.

关 键 词：机器视觉 能量 双目 标志点 立体匹配 

分 类 号：O438[机械工程—光学工程]