基于二维数据的快速极大似然抽样一致性算法  

Accelerated maximum likelihood estimate sampling consensus algorithm using two dementional data

作　　者：方磊[1] 王宏远[1] 徐帆[1] 田文[1] 

机构地区：[1]华中科技大学电子与信息工程系,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2008年第3期81-84,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

摘　　要：对传统估计两幅图像对应点的鲁棒性算法进行了分析,指出了其基于一维数据的局限性.在一般的随机抽样一致性策略的基础上,提出了基于二维数据的极大似然抽样一致性(MLESAC)算法,并用每组对应点的匹配点数与匹配强度指导抽样过程.在预检验模型参数评估随机抽样一致性策略的基础上,增加了后检验步骤及自动更新局外点比例的步骤,以此对MLESAC算法进行加速.在对简单场景与复杂场景的实验中,分别使对应点数量提高了26%和60%,从而改善了场景重建的质量.The limitation of traditional robust algorithm using one dementional data was pointed out after finding correspondences of two views using this algorithm was analyzed. General random sampling consensus （RANSAC） by using both feature count of each correspondence and match score to direct the sample process was discussed. A new maximum likelihood estimate sample consensus （MLESAC） algorithm based on two dementional data was proposed. By adding post-verify and automatic updating outlier fraction to preview model parameter evaluation RANSAC, MLESAC was accelerated. The experiments of the simple scene and complex scene show that the count of correspondence was increased by 26% and 60% respectively, thus the quality of the reconstruction was improved.

关 键 词：图像处理 对极几何 基础矩阵 随机抽样 极大似然抽样 一致性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]