基于高斯混合模型-地球移动距离的点集配准算法

Point set registration algorithm based on Gaussian mixture model-earth mover′s distance

提出一种基于高斯混合模型与地球移动距离的点集配准算法.将待配准的两个点集均表示为高斯混合模型,其中高斯分布的数量为点集中点的数量,每个高斯分布的均值为点的坐标值,方差为通过优化算法得到的优化值.在配准过程中通过优化两个高斯模型之间的地球移动距离来达到最佳匹配效果.该方法对点集配准中常见的噪声、外点、结构缺失等问题具有较强的鲁棒性.公共数据集与真实车辆平台上的实验表明该算法优于目前流行的点集配准算法.

A point set registration algorithm was presented based on Gaussian mixture model（GMM）and earth mover′s distance（EMD）.The two point sets were both denoted by GMM,where the number of Gaussian distributions was equal to the number of points in the point set.The mean of every Gaussian distribution was the locations of points and the covariance was optimized by a heuristic algorithm.The similarity of the two GMM was measured by EMD and optimized by an iterative manner.This method was robust to noise,outliers and missing partial structures.Both experiments on public data sets and real platform validate that the proposed method outperforms some state-of-the-art registration algorithms.