基于Jaccard距离与概念聚类的多模型估计

Multi-RANSAC、RHT等方法难以有效实现多模型估计。为此，提出一种基于模型聚类的多模型估计方法。将数据点描述为所属模型的倾向集，把倾向集间的Jaccard距离描述为数据点的一种属性，基于该属性使用改进的Cobweb算法进行聚类。该方法无需预知模型数目和参数变换，可有效克服漏检、交叉模型误检等情况。实验结果表明，该方法具有高效、高精度等优点，适用于消隐点检测、相机自标定等领域。

Multi-model Estimation Based on Jaccard Distance and Conceptual Clustering

Multi-RANSAC and RHT these methods are incapable to solve multi-models estimation effectually,and a multi-model estimation method with model-based clustering in conceptual space is proposed.Each data point is represented with a preference set of hypotheses models preferred by that point,and the Jaccard distance between two preference sets is described as a attribute of an data point,to perform a clustering operation using the improved Cobweb algorithm based on the attribute of the data points.Neither this method requires prior specification of the number of models,nor it necessitates parameters transformation,so that it can overcome missing detection and false detection of crossing models.Experimental results show the obvious effect and greater accuracy of the algorithm,thus can be used widely by vanishing point detection,self-calibration of camera,etc.