lukasvst
commented
1 year ago inc too small, break is a normal output of the algorithm during processing. I would recommend to set quiet=1 to suppress some of the commandline output.
Looking at your camera.txt, I think you need to change Pinhole to Equidistant, otherwise the algorithm will expect pre-rectified images. Note that even without IMU, the tracking should still be very accurate (and the pointcloud should look good), as long as you move the camera slowly and carefully. If this is not the case you first need to improve camera calibration before trying to add an IMU.
If VIO still does not initialize, it is best to have a look at the output lines starting with CoarseIMUInit normalized error:
DM-VIO will wait with initializing VIO until the variance is smaller than a threshold and will do a full reset if the error is larger than a threshold.
If the printed error is always large (even with good visual-only tracking), this suggests that IMU data / IMU calibration / IMU time synchronization is wrong; or that your IMU noise values are too small:
Also, I would highly recommend first trying out the system with the EuRoC and TUM-VI noise values (euroc.yaml / tumvi.yaml), rather than using your own noise values. Switching to own noise values is only recommended if you test it and it improves the results.
JensCallewaert
Hi, me again.
When I try to run the algorithm, it never switches to VIO and stays on Visual-only. This, of course, gives me bad results. I tried to look in the code to find how everything gets calculated but I don't seem to understand why it happens.
I get my calibration from Kalibr, i use EquiDistant because RadTan did not work when trying to calibrate the IMU with my camera.
If anyone understand what's going on here, I'd love to hear from you how I can fix this. Cheers!