一种多视图分层重构算法研究  

STUDY OF A HIERARCHICAL RECONSTRUCTION ALGORITHM BASED ON MULTIPLE VIEWS

作　　者：朱俊樵[1] 梁栋[2] 程志友[2] 

机构地区：[1]解放军电子工程学院,合肥230037 [2]安徽大学计算智能与信号处理教育部重点实验室,合肥230039

出　　处：《模式识别与人工智能》2003年第4期407-411,共5页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金(No.60143003);安徽省自然科学基金(No.01042206)

摘　　要：本文介绍了基于奇异值分解的射影重构算法的一般框架,以测量矩阵的秩为4作为约束,以仿射投影逼近透视投影,利用共轭梯度法估计射影深度,通过奇异值分解实现射影重构.利用共轭梯度法确定Kruppa方程中的未知比例因子,然后利用所确定的比例因子线性求解Kruppa方程.进而标定摄像机内参数.在摄像机内参数已知的情况下,求解一个满足欧氏重构条件的非奇异矩阵,然后通过此矩阵将射影重构变换为欧氏重构.实验结果表明所给出的算法是行之有效的。In this paper, the general framework of projective reconstruction based on SVD is introduced. Taking the measurement matrix rank 4 as the constraint, the affine projection is used to approximate perspective projection, the projective depths are iteratively estimated by using conjugate gradient method. The projective reconstruction is obtained by SVD of the measurement matrix. The conjugate gradient method is used to estimate the unknown scale factors in Kruppa's equations, then use the scale factors to solve Kruppa's equations linearly and calibrate the camera intrinsic parameters. In the case of the intrinsic parameters of camera are known, a non-singular matrix is evaluated, which satisfies the conditions of euclidean reconstruction, this matrix can transform projective reconstruction into euclidean reconstruction. The result indicates the algorithm is efficient.

关 键 词：计算机视觉 图像恢复 图像序列 共轭梯度法 射影重构 多视图分层重构算法 奇异值分解 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]