基于互信息的高精度双目视觉测距方法研究  

Improving Range-Measuring Accuracy of Stereo Vision Based on Mutual Information Registration

作　　者：任继昌[1] 杨晓东[2] 

机构地区：[1]海军潜艇学院学员二队,山东青岛266042 [2]海军潜艇学院航海教研室,山东青岛266042

出　　处：《控制工程》2015年第1期199-204,共6页Control Engineering of China

基　　金：海军装备部"十二五"预先研究项目(4010103010301-05)

摘　　要：在立体视觉测量中,为获得更高测量精度,拓展测距范围,最重要的是要获得精确的亚像素级视差。为此,引入图像配准领域常用的互信息理论并结合Powell优化算法实现亚像素级点匹配。首先,利用Harris角点探测器检测目标,并将得到的角点作为待匹配点,再采用最大互相关法对左图像进行匹配搜索确定匹配点。然后再对以左右匹配点为中心的一定大小的邻域图像进行插值,分别放大10倍和100倍。引入多分辨率配准策略,加快了图像配准速率。先采用互信息理论结合Powell优化算法对低分辨率图像进行配准,然后再在高分辨率图像上对第一次配准的结果进行细化,结合像素级匹配的整数视差可得到最终的亚像素级视差。实验结果表明,该方法能将视差精度提高到0.01像素,提高了测距精度。In order to improve the range-measuring accuracy in the stereo vision, a sub-pixel parallax is indispensable. Therefore, this paper introduces the Mutual Information theory which is usually used in the image registration, combined with Powell searching algorithm to realize the goal of sub-pixel point matching. At first, it uses Harris corner detector to find a most characteristic corner as candidate matching point. After that, it uses Most Cross Correlation matching rule to search the matching point. Then interpolation is carried out in a neighborhood for 10 and 100 times, whose center is the left and right matching point. It introduces multi-resolution method to accelerate the image registration speed. Firstly it registers the low-resolution image, after that to make the result more precision in the high-resolution image, combined with the integer-grade parallax, we can get the sub-pixel parallax. The result shows that the method used in the paper can improve the precision to 0.01 pixel.

关 键 词：亚像素 互信息 图像配准 立体匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]