I have 3 frames. I am trying to calculate the camera pose at third frame by iterative optimization. Using the first two frames, I've calculated 3D points by triangulation. Then, I projected some of these points onto the 3rd frame. Now, I can find the re-projection error but couldn't figure out how to optimize it with sparse solvers.
I'm prototyping with python. I decided to use the g2opy library for the whole project. I've already looked through the examples but there is no detailed information and no documentation. The examples are about full/sparse bundle adjustment and pose graph optimization.
Can you suggest any documentation / example ?
This technique is mostly used in SLAM, I suggest go to some SLAM paper, mostly Graph based SLAM.
Check this link, well described, you may find a solution to your problem.
http://wangxinliu.com/slam/optimization/research&study/g2o-1/
I suggest you follow https://github.com/uoip/g2opy and https://github.com/uoip/g2opy/blob/master/doc/g2o.tex
Thanks for the answers. I am trying to develop a monocular SLAM solution. The current problem is not the pose graph optimization. I'm just trying to estimate a single pose by optimizing reprojection error as in other pose estimation solutions.
Articles and examples are on the optimization of multiple poses and their relative measurements.
So, I tried to define reporj. error as a function and minimize it via scipy.optimization.minimization. It minimizes the error but, takes too long. I think it's definitely not the right way to go. I will calculate the Jacobian matrix and try classical Gauss-Newton optimization.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python