基于序贯概率及局部优化随机抽样一致性算法  

Fast and accurate RANSAC based on optimal sequential probability test and local optimization

作　　者：周骏[1,2] 陈雷霆[1] 刘启和[1] 李艳梅[1] 饶云波[1] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都610054 [2]西南大学计算机与信息科学学院,重庆400715

出　　处：《仪器仪表学报》2012年第9期2037-2044,共8页Chinese Journal of Scientific Instrument

基　　金：国家863计划重点项目(2007AA010407)资助

摘　　要：通过引入序贯概率检测及局部优化技术,给出了一种快速并具有较高精度估计的随机抽样一致性算法RANSAC(ran-dom sample consensus)。RANSAC算法是基于假设和检验的框架来实现。为了减少RANSAC算法的计算时间及提高计算精度,在模型的检验阶段,利用序贯概率检测技术,随机抽取少量的数据对模型的参数进行预检验。如果模型参数通过预检验,则进入后续的检验,否则这个模型参数不再参与后续全部数据的检验。在对模型参数估计时,使用局部优化方法,实现模型估计的精确度的提高。实验结果表明,此算法在速度和精确度方面与标准的RANSAC相比均取得了较大的提高。This paper presents a fast and accurate RANSAC （random sample consensus） algorithm basetl on optimal sequential probability test and local optimization technique. Basic RANSAC algorithm operates in a hypothesize-and-verify framework. In order to reduce the computing time of the RANSAC algorithm and improve its computing accuracy, at the model verification step, the optimal sequential probability test technique is used to randomly select a small number of data points to preview the model parameters. The remaining data points are evaluated only if the selected data points are all the inliers of the model, otherwise the model parameters do not take part in the following verification of all data points. When estimating model parameters, local optimization method is applied to increase the number of inliers found, which improve the accuracy of the model estimation. Theoretical analysis and experiment results show that the proposed algorithm can obtain better model estimation accuracy and faster computing speed compared with standard RANSAC algorithm.

关 键 词：随机抽样一致性算法 鲁棒性估计 模型估计 局部优化 

分 类 号：TP191.9[自动化与计算机技术—控制理论与控制工程]