一种三维刚体运动参数鲁棒估计的神经算法

A NEURAL ALGORITHM FOR ESTIMATING THE 3D RIGID BODIES' MOTION PARAMETERS

该文提出一种基于特征点匹配的刚体运动参数估计方法。在运动估计线性算法的基础上,文中利用全最小二乘(TLS)方法来进行求解,并建立次分量提取神经元来获得该全最小二乘解。基于测量数据中出格点(Outlier)的存在,我们在神经元的权值学习规则中引入鲁棒估计思想。实验结果表明,该方法能有效地克服出格点产生的误差,准确地估计出刚体的三维运动参数,比较令人满意。

This paper deals with estimating the rigid bodies' motion parameters from token correspondence between two perspective views. Total least square(TLS) method is used to solve the linear equations of motion estimation. Then a minor component analysis (MCA) neuron is constructed to get the TLS solution. Finally a outline resisting version of the learning rule is developed by using the statistical approach. Comparative experiments have been made and results show that our robust rules improve the performance significantly when outliers are presented.