多个三维刚体运动参数的鲁棒估计算法

A ROBUST ALGORITHM FOR ESTIMATING MULTIPLE 3-D RIGID MOTIONS

本文提出了一种估计多个三维刚体运动参数的鲁棒（ｒｏｂｕｓｔ）算法，可以处理包含高斯噪声和出格点（ｏｕｔｌｉｅｒ）的对应点数据．根据贝叶斯统计决策规则和蕴含在问题中的启发式规则，我们将运动参数估计问题转化为极大似然估计过程，实现部分模型拟合（ｐａｒｔｉａｌｍｏｄｅｌｆｉｔｔｉｎｇ）．因此，这种优化算法就是估计一组三维运动参数，使对应点数据最大限度地拟合似然函数，从而保证算法的鲁棒性．

This paper presents a robust estimation algorithm for multiple 3-D rigid motions,where the available point correspondences are corrupted by both the Gaussian noise and outliers. With aid of the Bayes decision rule and some heuristics implicit in the problem,the motion estimation is converted into the maximum likelihood (ML) estimation to maximally fit the unknown likelihood function only at observed point pairs.