Using Singal Fisheye camera loss tracking in indoor_dynamic dataset.

[FookSong]

Dear KIT team，
I have successfully run the example.
But， when I try to use only one fisheye camera of three to run the program in indoor_dynamic dataset，it always lost tracking .
I modified extractor.nFeatures: 1000 in the setting_indoor1.yaml as you wrote in the paper. And I'm sure the input images and calibration parameters are only belong to the choosen camera（I used cam2 in this exaple dataset）.

Is it right? Where am I setting wrong?

Hope you can give any reply, thank you！！！

[Hielamon]

I think that this situation for a single fisheye camera is really normal. Multi-fisheye cameras can capture the whole directional images and can attack this situation successfully.

[FookSong]

@Hielamon But the author said，It can also use Multicol-slam with single camera，I tried wide-angle camera（160） and fisheye camera（180）. But always loss tracking when I change the scenario. Do you know anyway to run with wide-angle camera or fisheye camera？

[xuqingwenkk]

I tried to use Multicol-slam with single fisheye camera before. Firstly, it initialized and worked fine, but It would lose tracking at some time and cannot relocalize any more. However, I think the problem is that the frequency of my image is too low. I am not sure whether your problem is that it cannot work with single camera or the performance of single camera is not good.

[FookSong]

@xuqingwenkk I restore the exam with 2-4 wide-angle(160) camera successfully. Unfortunately，I use 1 wide-angle camera，it loss tracking easily，even when I use the same data succeeded in more wide-angle camera. I still can not fix this problem. Maybe I should modify something to make it like ORB-SLAM. ~

[xuqingwenkk]

@FookSong Hmm, it seems that we met the same problem. Maybe it can be solved by improving the method of relocalization.

[drNoob13]

Have you calibrated your wide-fov fisheye camera before deployment?

[FookSong]

@TuanHo16 sure. I used the same way as the paper wrote.