摘要
首先在采用标签法预定位的基础上计算每个控制点的名义对应点 ,建立了控制点集的一种名义对应关系 ,然后将旋转变换矩阵和平移变换矩阵的计算描述为一个约束优化问题 ,根据优化理论中的 Kuhn-Tucker定理以及矩阵值函数的微商理论 ,利用奇异值分解导出了旋转变换矩阵和平移变换矩阵的计算公式 ,并通过迭代法计算视图间定位所需的坐标变换 ,从而快速方便地实现了多视定位。试验结果表明 ,本文提出的方法满足后续模型重建所需的精度要求。
The multi-view registration method is studied. On the basis of pre-registration with label method, a nominal correspondence of control point sets is created by computing the nominal corresponding point of each given control point. Then, the computation of the rotation matrix and translation matrix is formulated as a constrained optimization problem. According to the Kuhn-Tucker theorem and the properties on the derivative of the matrix-valued function, the formulae of rotation matrix and translation matrix are deduced by using singular value decomposition algorithm. Afterwards an iterative method is utilized to get the entire coordinate transformation of pair-wise views, thus the precise multi-view registration can be conveniently achieved. Experimental results demonstrate that the method is reliable,and can satisfy the precision need of subsequent model reconstruction.
出处
《数据采集与处理》
CSCD
2003年第3期346-350,共5页
Journal of Data Acquisition and Processing
基金
高等学校优秀青年教师教学科研奖励计划
江苏省青年科技基金 (BQ2 0 0 0 0 0 4)
航空基金 (0 1 H5 2 0 5 1 )资助项目
