基于归一化互相关的亚像素双目视觉匹配方法  

Sub-pixel Binocular Vision Image Matching Based on Normalized Correlation

作　　者：范新峰[1] 程远增[1] 付强[1] 

机构地区：[1]解放军军械工程学院电子与光学工程系,河北石家庄050003

出　　处：《探测与控制学报》2016年第3期71-74,共4页Journal of Detection & Control

基　　金：武器装备军内科研项目资助(装司2014551)

摘　　要：针对当前单一的区域匹配和特征匹配精度不高的问题,提出了基于归一化互相关的亚像素双目视觉匹配方法。该方法首先采用归一化互相关匹配获得目标图像的像素级最佳匹配点;然后根据互相关函数呈现出的特性,用改进的二次曲面拟合法拟合最佳匹配点周围区域,得到费马点,该点即为亚像素级的最佳匹配点;最后对两台相机采集的图像分别处理即可得到亚像素的匹配视差。实验结果表明,该匹配方法具有较高精度,能达到0.01像素。Aiming at the problem of low accuracy in a single regional matching or feature matching,an image matching method for binocular vision with sub-pixel accuracy based on normalized correlation was proposed.Firstly,the normalized correlation matching was used to get the optimal matching point with pixel accuracy.Then the correlation peak was fitted according to the feature of correlation function by method of improved quadratic surface fitting,and the Fermat point was gained,which was the optimal matching point with sub-pixel accuracy.Finally,the matching parallax with sub-pixel accuracy was calculated by photos of two cameras.Experimental results showed this method had high sub-pixel accuracy to 0.01 pixel.

关 键 词：归一化互相关 二次曲面拟合 图像匹配 亚像素 双目视觉 

分 类 号：TN911.73[电子电信—通信与信息系统]