计算机视觉中强鲁棒性的遗传一致性估计

Genetic Consistency Estimation for Strong Robustness in Computer Vision

用于估计计算机视觉模型的传统鲁棒算法均存在估计精度和稳定性不高等问题。为此,结合遗传算法的全局最优性及几何模型估计的特殊性,提出一种强鲁棒性的遗传一致性估计算法,以估计各种误差和错误概率下的计算机视觉几何模型。仿真实验结果表明,相比于RANSAC、MAPSAC、MLESAC等鲁棒算法,该算法在估计精度和鲁棒性方面性能更优。

Traditional robust algorithms for estimating models in computer vision are inevitably affected by errors and outliers in the provided data,which makes estimation precision and stability low.This paper presents a Genetic Consistency Estimator（GCE） with strong robustness,combining global optimality of genetic algorithm and specialty of model estimation,to estimate geometric models in computer vision in a variety of errors and outliers probabilities.Experimental results prove GCE has greater precision and robustness compared with RANSAC,MAPSAC,MLESAC and other robust algorithms.