高级检索

    耿冬冬, 罗娜. 一种基于多邻域非线性扩散的动态规划全局立体匹配算法[J]. 华东理工大学学报(自然科学版), 2017, (5): 677-683. DOI: 10.14135/j.cnki.1006-3080.2017.05.012
    引用本文: 耿冬冬, 罗娜. 一种基于多邻域非线性扩散的动态规划全局立体匹配算法[J]. 华东理工大学学报(自然科学版), 2017, (5): 677-683. DOI: 10.14135/j.cnki.1006-3080.2017.05.012
    GENG Dong-dong, LUO Na. A Dynamic Programming Global Stereo Matching Algorithm Based on Multiple Neighbors' Nonlinear Diffusion[J]. Journal of East China University of Science and Technology, 2017, (5): 677-683. DOI: 10.14135/j.cnki.1006-3080.2017.05.012
    Citation: GENG Dong-dong, LUO Na. A Dynamic Programming Global Stereo Matching Algorithm Based on Multiple Neighbors' Nonlinear Diffusion[J]. Journal of East China University of Science and Technology, 2017, (5): 677-683. DOI: 10.14135/j.cnki.1006-3080.2017.05.012

    一种基于多邻域非线性扩散的动态规划全局立体匹配算法

    A Dynamic Programming Global Stereo Matching Algorithm Based on Multiple Neighbors' Nonlinear Diffusion

    • 摘要: 双目立体视觉匹配通过两幅具有一定视差的图像获得精确、稠密的视差图。为了解决动态规划立体匹配算法橫条纹瑕疵以及精度低的问题,提出了一种基于多邻域非线性扩散的立体匹配算法。该算法采用AD测度函数构建视差空间,根据行列像素之间的约束关系,基于非线性扩散的代价聚合方法,通过图像边缘的动态优化寻求全局能量函数最优值得到稠密视差图。在Middlebury测试集上的实验结果表明,该算法的平均误匹配率为5.60%,相比IIDP动态规划全局匹配算法,精度提高了39.9%,有效地解决了横向条纹问题,改善了边缘模糊情况,且提升了算法的稳定性。与其他全局匹配算法相比,本文算法误匹配率降低了38.2%,在图像参数的11个指标中有9项指标排名第1。

       

      Abstract: Binocular stereo matching can obtain the accuracy and dense disparity map by comparing two images.However,the utilization of dynamic programming algorithms may result in some shortcomings,such as stripe-like and low accuracy.Aiming these problems,this paper proposes a new stereo matching algorithm based on multiple neighbors' nonlinear diffusion.Firstly,absolute difference test method is used to build disparity space image in raw costs computation period.And then,according to the constraint relation between rows and columns,multiple neighbors' nonlinear diffusion of costs aggregation is proposed to improve the global costs function.Finally,dense disparity maps during the global optimization process are obtained by the edges-optimized DP optimization.The experiment results via Middlebury test images show that the proposed algorithm attains the average PBM 5.60% and raises the accuracy 39.9% than ⅡDP.Moreover,the problem of stripe-like is well solved and the edge-blurring is also improved.Compared with other global matching methods,the proposed algorithm reduces PBM by 38.2% and has 9 of 11 indexes to rank the first.

       

    /

    返回文章
    返回