含局部空间约束的t分布混合模型的点集配准

摘要

基于高斯混合模型(Gaussian mixture model, GMM)的点集非刚性配准算法易受重尾点和异常点影响,提出含局部空间约束的t 分布混合模型的点集非刚性配准算法。通过期望最大化(Expectation maximization, EM)框架将高斯混合模型推广为t分布混合模型；把Dirichlet 分布作为浮动点的先验权重,并构造含局部空间约束性质的Dirichlet 分布参数。使用EM算法获得配准参数的闭合解；计算浮动点的自由度,改变其概率密度分布,避免异常点水平估计误差。实验表明,本文提出的配准算法具有配准误差小、鲁棒性好、抗干扰能力强等优点。%A robust non-rigid registration framework using the student0s-t mixture model with spatial constraints is proposed in this paper. The Gaussian mixture model which is vulnerable to outliers and data longer than normal tails is a special case of the student0s-t mixture model in theory. The Dirichlet distribution is used as a prior distribution to reduce the impact of outliers. The Dirichlet parameter set with spatial constrains is structured to incorporate the spatial information into the decision process. The closed form solution of the parameter set of the student0s-t mixture model is solved by re-parameterizing the student0s-t mixture model in the expectation maximization (EM) algorithm. The degree of freedom of each moving point is calculated to change the probability density to reduce the registration error. It can also avoid estimating the outlier level of data sets that may bring additional error. The experiments showed that this non-rigid registration algorithm has features of high-accuracy and good robustness compared to other point set registration approaches.