Mono-Inertial tracking continues even after bad results

[Pallav1299]

I am using ORB-SLAM3 in MONO-INERTIAL mode with Realsense D435i.

Problem: Sometimes tracking degrades due to sudden motion and the tracking tends to drift further away even after I make the device static.

Expectation: I expect new map initialization / relocalization / loop closure to happen in such situation.

Though I am yet to go through the code, I suspect that the issue might be in the mechanism to understand the tracking performance.

Any views regarding the same?

[Pallav1299]

Hi, I used the rolling shutter RGB sensor with the inbuilt IMU on the D435i and was able to achieve decent results with ORB-SLAM3. I must admit that the performance starts to degrades very much at times of fast motion and doesn't recover at all. Although the reason might be due to the rolling shutter in the RGB camera as mentioned HERE.

Keeping that aside, while the tracking keeps on degrading there isn't any new map initialization. That forces me to think about the failure detection module and multi-map merging applications in ORB-SLAM 3, which should initialize a new map in such a scenario. I couldn't find anything similar mentioned in the ORB-SLAM3 paper. Is there any failure detection happening in ORB-SLAM3?

I could find reference to the failure detection module in the VINS-Mono research paper.

[mateosss]

I have encountered similar issues with both mono-inertial and stereo-inertial modes on a D455. Usually, it starts happening after a couple of minutes.

[Abner0907]

@Pallav1299 I am also working on d435i in mono-inertial, But it can not detected the orb features and can not create the map, can you show me your d435i's configure file? and have you do some changes in the code inside? Thank you very much

[songxuekun]

5上遇到了单惯性和立

do you solve the problem?