Limitations of relocalization

[ymd-stella]

The conditions for successful relocalization are difficult for non-specialists to understand. In this issue, we will consider adding documentation, logs, and visualization.

Documentation

Close camera position and camera angle are relatively easy conditions to understand. This is essentially the presence of enough landmarks that can be matched with the features of the frame. In Equirectangular, the relocalization succeeds regardless of the yaw angle of the camera. For models other than Equirectangular, relocalization fails if the yaw angle from the keyframe is significantly different even if the position from the keyframe is the same. See #79 (comment).

Logs

If candidates are empty, or if candidates are rejected, the reason for the rejection should be shown.

Visualization

Since the position and orientation of the candidate keyframes for relocalization are important, it is necessary to be able to see which ones are the candidate keyframes.

TODO

• Add documentation
  • Describes the rough flow of the relocalization process #142
  • Describes how relocalization can fail if the angle from the candidate keyframe is too large #143
• Add logs
  • Add debug level log to relocalizer.cc #141
• Update visualization
  • Visualize candidate keyframes for relocalization #140

[jeffin07]

@ymd-stella Hi can i work on this issue?currently I'm exploring slam so contributing will help me to understand.I see documentation and other task which will be good for a beginner that's why i asked

[ymd-stella]

I have divided the task into multiple issues. If you have any questions, please ask them in each issue.

If you have never run relocalization before, I suggest you try it. For example, if you run localization on a map created by running slam, relocalization will be performed during initialization. See https://openvslam-community.readthedocs.io/en/latest/simple_tutorial.html. In addition, it is better to make sure that relocalization is also performed when tracking fails.