立方体姿态估计

最新推荐文章于 2024-06-05 16:51:29 发布

MusingHead

最新推荐文章于 2024-06-05 16:51:29 发布

阅读量678

点赞数

本文链接：https://blog.csdn.net/qq_37285783/article/details/104236008

版权

本文介绍了如何利用Ceres-solver最小化立方体的重投影误差和光度损失。通过旋转矢量和局部坐标系理解aruco标记，然后构建并优化6个参数（旋转和平移）。实验结果显示，aruco库提供的初始估计已经接近局部最优解，且在加入噪声后优化结果与未扰动时相近。同时，探讨了光度损失的计算和优化过程，尽管某些情况下未达到理想收敛状态。

摘要由CSDN通过智能技术生成

使用Ceres-solver最小化reprojection error和photometric loss

install (on Mac + Clion)

brew install ceres-solver
brew install opencv

CMakeList.txt

find_package(Ceres REQUIRED)
include_directories(${CERES_INCLUDE_DIRS})
target_link_libraries(helloworld ${CERES_LIBRARIES})

Rotation Vector

Three dimensions
Direction is the axis and magnitude is the angle in radians (counter clockwise, axis centered)
Use Rodrigues’ rotation formula to calculate (k is the unit vector of the rotation axis):

$v_{rot} = v\cos\theta + (k \times v)\sin\theta + k(k \cdot v)(1-\cos\theta)$

Local Coordinate of an Aruco Marker

A detected marker: 4 corners, [top-left, top-right, down-right, down-left]
Let length of one square side be L, corners’ coordinate:
1. Top-left: [-L/2, L/2, 0]
2. Top-right: [L/2, L/2, 0]
3. Down-right: [L/2, -L/2, 0]
4. Down-left: [-L/2, -L/2, 0]

Minimize Reprojection Error

（Ceres-solver优化时，需要计算loss function对各参数的偏导的解析解，因此在构造reprojection error时不能使用外部的库，只能使用直接的代数计算和ceres-solver提供的函数）

Optimize 6 parameters:

R: 3 paramaters for rotation using Rodrigues’ rotation formula
t: 3 parameters for translation

Convert from world to camera coordinate, (P: point in camera coordinate, X: point in world coordinate):
$\tag{1}$
Perspective projection:
$\tag{2}$
Radial distortion:
$k_1 * ||p||^2 + k_2 * ||p||^4 \tag{3}$