Can not correct the loop? about orb_slam3

[ccamposm]

Hi @liangsongLee.
If both candidate images for loop-closing have a similar point of view (ORB descriptors are likely to be similar) and there exist enough map points, there shouldn't be any problem. Is that the case?

[liangsongLee]

@ccamposm Hi, I find the problem is here:

ORB_SLAM3/src/LoopClosing.cc

Line 687 in adc4062

if(bConverge)

bConverge is always false, so it can not do the following process in loop closing.
The reason for this is here:

ORB_SLAM3/src/Sim3Solver.cc

Line 277 in adc4062

if(mnInliersi>mRansacMinInliers)

mnInliersi>mRansacMinInliersis false. so bConverge is false. However, when I use pinhole camera, this situation will not happen.Is it caused by the fisheye camera?

[ccamposm]

You should check your calibration for fisheye, since loopclosing is similarly performed for any camera model. Reopen if problem persists.

[ouyangandy]

@liangsongLee 请问您解决了这个问题吗？我使用廉价的单目鱼眼相机，进行标定后也是这个结果，始终无法检测到闭环，看到了您的分析，但不知道该如何处理。

[liangsongLee]

You should check your calibration for fisheye, since loopclosing is similarly performed for any camera model. Reopen if problem persists.

Hi, I tested TUM VI open datasets with loops, using their published calibration params of fisheye, but it can not detect the loop either. So I guess it is not the problem of my calibration.

[liangsongLee]

@liangsongLee 请问您解决了这个问题吗？我使用廉价的单目鱼眼相机，进行标定后也是这个结果，始终无法检测到闭环，看到了您的分析，但不知道该如何处理。

没有解决，我试了tum鱼眼公开数据集，用了他们公开的鱼眼标定参数，也不能闭环，针孔相机有时候也不能闭环。

[ouyangandy]

@liangsongLee 请问您解决了这个问题吗？我使用廉价的单目鱼眼相机，进行标定后也是这个结果，始终无法检测到闭环，看到了您的分析，但不知道该如何处理。

没有解决，我试了tum鱼眼公开数据集，用了他们公开的鱼眼标定参数，也不能闭环，针孔相机有时候也不能闭环。

测试一下最新版本库的代码。我更新了一下代码，今天多尝试了一组数据，发现能闭环了，具有一定的偶然性。使用的是单目鱼眼相机，且用的是自己买的廉价相机，使用kalibr进行标定的。