基于采样优化的随机抽取一致性算法

Fast and Accurate RANSAC Based on Sampling Optimization

为了提高随机抽取一致性算法(RANSAC)的效率和精度,提出了一种基于采样优化的随机抽取一致性算法。首先通过匹配点的相似性度量计算匹配点先验概率,根据先验概率随机抽取最小子集估计模型,在全部数据上检验模型,依次迭代找到次优模型;然后以次优模型对应的内点集作为采样的初始集,随机抽取最小子集估计模型,并在全部数据上检验模型,若模型更好则更新采样初始集,依次迭代找到最优模型;最后,选择最优模型获得符合该模型的内点和最终的模型参数。选取多对不同变换的图像作为实验数据,从算法运行效率和模型精确度两方面对算法进行了测试实验。实验数据表明,本文算法的迭代次数约为标准RANSAC算法的20%,运行时间约为标准RANSAC算法的25%,标准误差降低了30%左右。本文算法充分利用了匹配点的先验知识和模型检验结果对采样模式进行优化,算法的运行效率和精度都有较大提高。

This paper presents a fast and accurate Random Sample Consensus（ RANSAC） algorithm based on sampling optimization. Firstly,the prior probability of the matching points is calculated by the similarity measurement of matching points,and the minimum subset for model fitting is selected randomly according to the prior probability,which is tested on all the data,until the suboptimal model is found through iteration. Then,the interior point set corresponding to the suboptimal model is used as the initial set for sampling and the minimum subset of the model is randomly extracted and tested on all the data. If the model is better,then the initial set is updated,and the optimal model is found through iteration. Finally,the optimal model is selected,and the interior point and the final model parameters are obtained. The images with different changes are selected as the experimental data,and the algorithm is tested on both the algorithm operation efficiency and the model precision.The experimental data show that the number of iterations in this algorithm is about 20%,and its running time is about 25% of the standard RANSAC algorithm,and the standard square root error is reduced by about 30%. This paper makes full use of the prior knowledge of the matching point and the results of the model test to optimize the sampling mode,so that the operation efficiency and precision of the algorithm are greatly improved.