Pose graph optimiaztion with landmark detection?

[szx0112]

Hi, open3d is a great tool for rgbd-based 3d reconstruction. Currently, the pose graph contain the odometry and loop closure nodes. However, in my application (survey a tunnel through a single pass), loop closure is not existed. And we known camera poses estimated only through odometry will drift.

An alternative solution is to use landmarks. The tunnel contains connections at fixed intervals (like every 10 meters), and we can detect them in the images. I am wondering if it is possible to add those landmarks into the pose graph in order to improve the camera pose estimation results? If yes, how can I do it?

Any comment is appreciated.

[theNded]

Our pose graph is limited and does not support landmarks. I would like to redirect you to

• Ceres: http://ceres-solver.org/
• g2o: https://github.com/RainerKuemmerle/g2o
• GTSAM: https://github.com/borglab/gtsam