openMVG----sfm

sfm模块处理SfM相关数据的存储和解决SfM问题的方法（相机位姿估计，结构三角重构，捆集调整）

一个通用的SfM数据容器

SfM_Data类包含了SfM的相关数据

• 一系列View
  • the used images
• 一系列相机外参
  • the camera poses
• 一系列相机内参
  • the camera internal projection parameters
• 点云，即结构
  • the collection of landmark (3D points associated with 2d view observations)

struct SfM_Data
{
  /// Considered views
  Views views;
  /// Considered poses (indexed by view.id_pose)
  Poses poses;
  /// Considered camera intrinsics (indexed by view.id_cam)
  Intrinsics intrinsics;
  /// Structure (3D points with their 2D observations)
  Landmarks structure;
  
  // ...
}

View concept

包含和图像相关的信息：

• image filename
• id_view (must be unique)
• id_pose
• id_intrinsic
• image size
  因为采用了id_*，所以可以共享位置和内参很方便
  该类型是抽象的。可以在此基础上自定义新的类型

Camera Poses concept

存储全局的Ri,Ti

Camera Intrinsic concept

该类型是抽象的。可以在此基础上自定义新的类型。可共享

Structure/Landmarks concept

3D点云

SfM_Data cleaning

去除外点的通用接口：

• 给定的像素残差
• 沿着track的最小角度

Triangulation

• 一般方法(blind):
  • Triangulate tracks using all observations,
  • Inlier/Outlier classification is done with a cheirality test,
• A robust method:
  • Triangulate tracks using a RANSAC scheme,
  • Check cheirality and a pixel residual error.

Non linear refinement, Bundle Adjustment

openMVG提供通用的BA框架，可以对以下参数进行改进或保持不变：

• internal orientation parameters (intrinsics: camera projection model),
• external orientation parameters (extrinsics: camera poses),
• structure (3D points).

SfM_Data sfm_data;
// initialize the data
// ...

const double dResidual_before = RMSE(sfm_data);

// Bundle adjustement over all the parameters:
std::shared_ptr<Bundle_Adjustment> ba_object = std::make_shared<Bundle_Adjustment_Ceres>();
ba_object->Adjust(sfm_data);

const double dResidual_after = RMSE(sfm_data);