Application of VINS-Mono on Underwater Dataset

[Anejdj]

Hi everyone, I’m currently applying VINS-Mono to an underwater dataset and have encountered a few issues that I’d like to ask for help with:

1、The estimated trajectory is significantly shorter than the ground [truth.]

2、I noticed that the tracked feature points (those white dots in the image) are very close to the camera — does this suggest that the scale was estimated incorrectly during initialization?

3、I also observed that after each turn in the trajectory (the ground truth is a looping rectangle), the estimated speed drops significantly, to the point where the system ends up almost spinning in place near the end.

Any help or insights would be greatly appreciated. Thanks in advance!

[brucezhcw]

The camera focal length F depends on the refractive indices of both the lens material and the surrounding medium.

That is to say, the camera focal length must be adjusted in underwater environment

have you noticed that？

[yangato]

Your camera distortion parameters appear to be incorrect. The wide angle lense you used will cause the straight lines to be round. You will often see this manifest in long straight trajectories as a bowing or bending.

The speed drop may be do to a loss of scale due to poor tracking of objects on turns. This often gets progressively worse with more degenerate motions.