嵌入正交权值神经网络在摄像机内外参数标定中的应用  

Application of neural network with embedded orthogonal weight to calibration of camera's intrinsic and extrinsic parameters

作　　者：葛动元[1,2] 姚锡凡[1] 向文江[2] 

机构地区：[1]华南理工大学机械与汽车工程学院,广东广州510640 [2]邵阳学院机械与能源工程系,湖南邵阳422004

出　　处：《光学精密工程》2011年第11期2782-2790,共9页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.51175187);国家863高技术研究发展计划资助项目(No.2007AA04Z111);湖南省自然科学基金资助项目(No.09JJ6092);湖南省教育厅资助科研项目(No.09B092)

摘　　要：针对机器视觉在某些应用场合对摄像机内外参数的需要,提出了一种基于内嵌正交权值矩阵神经网络的摄像机标定方法。首先,使设计的神经网络的权值分别与摄像机的外参数和内参数相对应,即使得所设计的神经网络与摄像机的物理模型一致。正交权值矩阵的生成在迭代中相当于遗传算法的上一次变异,系统的性能指标为由网络输出组成的矢量与对应特征点投影于图像平面的齐次坐标差值2-范数的平方。同时引进混合遗传-模拟退火算法,使得系统在达到平衡状态时可根据神经网络的权值完成摄像机内外参数的标定。实验结果表明,该方法具有较好的鲁棒性,且标定精度高,算法简洁,是摄像机内外参数标定的有效解决方案。According to the intrinsic and extrinsic parameters of a camera to be needed in some applications for machine vision,a method based on the neural network with embedded orthogonal weight matrix is proposed for camera calibration.Firstly,the neural network with the embedded orthogonal weight matrix is structured,whose weights are corresponding to the camera extrinsic parameters and intrinsic parameters.Thus,the structured neural network coincides with the model of camera.The generation of orthogonal weight matrix is served as the last mutation operator in the iteration of genetic algorithm,and the performance index is the square of 2-norm of the difference between vector consisting of network＇s outputs and homogeneous coordinate of corresponding feature point projected in image plane.Meanwhile,a hybrid genetic-simulation annealing algorithm is introduced into the solving-programming.When the system comes to the equilibrium,the intrinsic and extrinsic parameters of the camera can be obtained in the light of network＇s weights.The simulated experiments illustrate that the proposed algorithm is robust,and has the advantages of high calibration precision,simplicity and clarity according to real experiments.It provides an effective solution scheme for camera calibration of intrinsic and extrinsic parameters.

关 键 词：机器视觉 摄像机标定 正交神经网络 遗传退火算法 哈明距离 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]