移动视觉测量中基于编码网络的特征点匹配方法研究  

Study on feature points matching method based on coding network for portable visual metrology

作　　者：隆昌宇[1] 邾继贵[1] 孙博[1] 王伟[1] 林嘉睿[1] 

机构地区：[1]天津大学精密测试技术及仪器国家重点实验室,天津300072

出　　处：《光电子．激光》2014年第5期937-946,共10页Journal of Optoelectronics·Laser

基　　金：国家高技术研究发展计划"863计划"(2012AA041205)资助项目

摘　　要：为解决移动视觉测量中多图像间利用极线约束进行特征点匹配时基本矩阵求解精度很难进一步提高问题,提出一种基于网络模型的间接基本矩阵求解方法。首先,利用编码点的自动无误匹配信息建立空间交会共线数学模型;其次,经过优化数学模型建立精密编码网络,并对相机内部参数和测量站位外部方位参数进行了优化;再次,利用基本矩阵与这些参数之间的关系间接确立基本矩阵;最后,通过精确恢复的极几何模型完成不同图像间特征点的精确匹配。与V-star结果进行比对,验证了编码网络的建立误差平均值为0.051 85mm,均方差为0.020 89mm;与两种经典方法在匹配正确率和极几何恢复精确度方面进行了比对,证明了本文可以精确恢复极几何模型并能提高特征点匹配率。Portable visual metrology works with multiple uncalibrated images and image feature points need to be matched by epipolar geometry. Due to the existence of image distortion, fundamental matrix can＇t be accurately solved. In order to resolve this problem,a novel indirect fundamental matrix solving method based on coding network geometry is proposed. In this method, spatial interaction collinear math- ematical model is established by using automatic error-free correspondence of coding points firstly. Sec- ondly, after being optimized,the strong coding network geometry is built accurately. Meanwhile, the in- ternal and external orientation parameters of each station are optimized accurately. Thirdly, the funda- mental matrix can be solved indirectly with the relationship between fundamental matrix and these pa- rameters. Finally,feature points from multi-view can be matched correctly by using the recovered epipo- lar geometry. Experiments are done for the verification of this method. Compared with v-star, the average error and mean square error of establishing coding network are 0. 05185 mm and 0. 020 89 ram,respec- tively. Besides, the right matching rate and epipolar geometry recovering accuracy of this method are compared with those of two classic methods,including iterative levenberg-marquardt method and robust LMedS method. Extensive experimental results prove that this method can recover the epipolar geometry accurately and find more matches.

关 键 词：特征点匹配 基本矩阵 移动视觉 编码网络 共线方程 

分 类 号：TH741[机械工程—光学工程]