欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi

程序员文章站 2022-03-29 19:25:14
...

【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matching Score

0、 basic Information

Pons J P , Keriven R , Faugeras O . Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matching Score[J]. International Journal of Computer Vision, 2007, 72(2):179-193.

1、method overview

  1. 核心:
    基于全局(all image pairs i,j) 计算image_i 与image_i’(predicted by image j)之间的相似性 M,并基于相似性与regularization 构造能量函数,并采用梯度下降的方法( 简化: E只考虑一阶导),迭代求解,得到使E最小化的surface
  2. surface 的最终迭代公式(推导见下):
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi

2、similarity measure

2.1、NCC

  1. 基本原理:见参考链接????
  2. 本文 :Gaussian windowssize (fix)– non-distortion】
  • 数学表达:
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi

  • 梯度、微分表示:
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi
注意: alpha bata gamma 关于x 的迭代过程中,变化较小,10 次迭代才更新

2.2、MI( mutual information)

  1. 实质: statistical dependencies –联合概率分布、边缘概率分布
  • 数学表达:
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi

  • 梯度、微分表示:
    【论文阅读】19-Multi-View Stereo Reconstruction and Scene Flow Estimation with a Global Image-Based Matchi
注意:ikseluo 10次迭代才更新

3、regularization

  1. 目的: preserve shape motion discontinued
  2. User-difined
  3. Basic-smoothing:the area of the surface----mean curvature(曲率) motion
数学表达 : nameda * H(surface mean curvation)

4、details

  1. 图像预处理:smoothing && subsample
  2. Pix -只考虑visible( distortion && occlusion removed)

5、implementation

  1. Level set – 数值稳定,处理拓扑变化
  2. OpenGL depth buffer – visibility
  3. Texture mapping – reprojection — 未细究
  4. Shadow-mapping – semi-occluded — 涉及occlusion 检测,未细究
  5. 参数设置

6、result

  1. Camera pair ( neighboring)
  2. Quantity指标
Difference ( result && true )  / true  单位:volume (体积)