Multi session mapping w/o "Localization Mode" trick between sessions

[HTDcris]

Hi Matlabbe,
I followed your tutorial carefully:
https://github.com/introlab/rtabmap/wiki/Multi-Session-Mapping-with-RTAB-Map-Tango
I used the files of three capture sessions of a small natural cave. Each scan overlaps in several points the previous, but unfortunately I can not get a single map at the end of the processing. Do you think it might be due to the fact that I have not used the "Localization Mode" trick between sessions? I took these scans weeks before the publication of your tutorial...

The 3 database files are here:
https://drive.google.com/drive/folders/0B3xmLzdcR2kENHBJRU1rbHJjRHc?usp=sharing

An early draft-model achieved with manual alignment of the three point clouds can be seen here:
https://skfb.ly/668PL

Do you think that a single database is still achievable perhaps changing some settings? Or there is no hope at all?

Thank you so much

Cris

[matlabbe]

Hi Cris,

Using database viewer, can you point out which images (Id) would correspond? I tried manually browsing all images and I didn't find corresponding images between two sessions. Note that overlapping point clouds/images should be taken from the same point of view to be able to find a loop closure. Unfortunately with these scans, it is indeed hard (may not possible) to get a single database with all maps merged automatically.

cheers,
Mathieu

[HTDcris]

I think that actually the scans overlap portions of the same objects but from different angles. A human can find correspondences between the two point clouds, but if the software requires two identical pictures or POVs to do the same... I understand why it fails.

I'm pretty sure that at least between 3 scans there are some identical pictures but I'm not able to retrive them to be honest. I'll pay more attention next time.

Thanks

Cris