基于修正最小平方中值定理的鲁棒运动估计

Robust Motion Estimation Based on the Modified LMedS

基于修正的最小平方中值定理（ＬＭｅｄＳ），本文提出一种由３Ｄ特征点空间位置估计运动参数的鲁棒方法。首先由ＬＭｅｄＳ给出初始的运动参数估计，然后采用迭代加权估计方法重新计算运动参数，其中一种Ｔｕｋｅｙ权和Ｈｕｂｅｒ权的混合权函数代替原ＬＭｅｄＳ的二分权函数。该算法减轻了当信噪比较低的情况下删除一些出格点的困难，故可得到更好的估计精度。计算机模拟表明其性能令人满意。

Based on the modified least median of squares (LMedS)regression,a robust approach of motion estimation from the space position vectors of 3-D feature points is put forward.First initial estimation is obtained by the primary LMedS,then we re-estimate motion parameters with an iterated reweighted estimator, in which a hybrid weight function of Huber weight and Tukey weight takes place of the dichotomy weight in the primary LMedS.This algorithm alleviates the difficulty of deleting some outliers while SNR is too low,so can get better accuracy of estimation. Computer simulations show that its performance is satisfactory.