Hard to get Loop Hypothesis success in a specific corner

[Gary_Allison]

Hey, I am new to the RTab-Map and trying to use RTabMap to do some Point Cloud Mapping. I am using D455 with a T265 Intel Deep Camera.

Due to the camera has only 5-6 m limitation. I understand that I need to keep mapping along with certain items like a wall or something else. But when I was trying to map a corner, which has a lot of references. It will always give me some bad results.

I

I am using RTabMap Windows Version.

I also attached my config file. rtabmap.ini

[matlabbe]

Not sure exactly what is the problem. Do you mean loop closures are always rejected even if there are a lot of features? If you can share the database, it could help to see the problem.

[Gary_Allison]

Thanks for the reply. I have attached a database here. It has the same problem that It seems a very similar picture, but Loop Detection rejected it.

https://1drv.ms/u/s!Ahd_lIAtyq8NnYEGqdfS2JOALKMYXg?e=4lVhoV

[matlabbe]

In the database, the big loop closure has been detected. Before:

After:

In the graph view, we can see he red links between the first and second traversal:


Can you point me which image IDs in rtabmap-databaseViewer the loop closure has been rejected. There should be a warning message in terminal telling why the loop closure has been rejected.

[Gary_Allison]

Appreciate the support. After following the instruction of "Change Parameter", the result is getting much better. However, I still have some issues but it is not related to the topic. I will start a new thread.