-
Notifications
You must be signed in to change notification settings - Fork 805
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Failing Localization with Obstructions Present while still #1444
Comments
Are you using a stereo or RGB-D camera? If a depth image is used, by default features are not extracted when there is no depth. You may disable this by setting Note that another way to debug localization is to backup your map database, then restart in localization mode with |
@matlabbe so I have set the location images with a dummy pose and was able to open the databse in rtabmap-databaseViewer with the location images. I have ran location, against a map created with superpoint and superglue (complete config end of the comment). The matches I can see in rtabmap-databaseViewer are much less than the onews I obtain when I run superpoint+superglue from the superglue repository. Matches shown in databaseViewer: Matches running superpoint+superglue using demo_superglue.py (config identical): map config: |
To verify that superglue is giving similar results on both cases, you can try
You can post the log here with |
I am experiencing difficulties with RTAB-Map's localization and loop closure capabilities in scenarios where there are obstructions (specifically, people) between the camera and previously mapped areas, but there are still important scene features visible. In this scenario the camera subject is still. I don't experience these issues if the people are not obstructing the view.
Observations:
Scenario with occlusion, with bad localization

There are some good keypoint matches, despite some bad ones.
Image on the right is an rgb exported from the map at a close position. Using distance threshold 0.7
Scenario without occlusion with good localization

Good keypoint matches, but again not many.Image on the left is from an rgb exported from the map
Expected Behavior:
I would expect RTAB-Map to be more robust to partial occlusions, especially when feature matching (as demonstrated by external SIFT) is still possible. Ideally, it should be able to utilize the available matches to achieve at least a coarse localization or loop closure, even if the accuracy is slightly reduced due to the obstructions.
Question:
What factors within RTAB-Map's processing could be contributing to these failures in the presence of partial occlusions, despite the apparent availability of matching features? Are there specific parameters or configurations that could be adjusted to improve robustness in such scenarios?
The text was updated successfully, but these errors were encountered: