Open BearrGod opened 3 months ago
Do you have an example bag with your configuration files I could test with?
The logic to append new SLAM features should be pretty independent of the number of cameras. There might be a bias to selecting new SLAM features in smaller camera ids, but it seems you don't have any at all after using the third camera. https://github.com/rpng/open_vins/blob/e274c656324f84b9f2730eacd0cf54a6f8b74bb8/ov_msckf/src/core/VioManager.cpp#L440-L453
Hello @goldbattle. I have a dataset here that you could use to test my 3 cameras setup. I changed my extrinsics in the mean time (no overlap in field of view) so no stereo constrains. Here is the bag file. My config files are attached here config3camsbas.zip . Additionnaly i've done a few test that might point in the right direction. When reading the code , i realized that when there are two ameras, they are automatically synchronized (sync_pol ) . So i did the same for the 3 cameras (meaning 3 images and one timestamp in the camera measurement) and the slam starts working. But i wasn't satisfied with it since all the slam points tend to be on my camera 0 . On the other hand i tried to print out the _amount_toadd and _validamount in the piece of code that adds new slam features and surptisingly the _validamount is always equal to zero when using 3 cameras with my setup. I hope ths helps out. Thank you.
Hello everyone. I'm using openvins with 3 cameras (150 degrees HFOV) at 45 deg from one another making a decent overlapping field of view. But when i try to use the 3 cameras, the slam part does not work ( no slam points published) which is fine if i keep moving the platform around, but the moment a stay stationary the estimated pose drifts away. But when i activate only 2 cameras (max_cameras:=2), the slam works. Any help on the matter ? Here is my config file :