Remarks about azure Kinect and question about recommended settings

[Iga]

Hello, I've been doing some tests with Azure Kinect and standalone Rtab-map.
As I am a beginner and not a programmer I also followed the settings in post:

http://official-rtab-map-forum.206.s1.nabble.com/Kinect-For-Azure-L515-ICP-lighting-invariant-mapping-td7187.html 

To my understanding, in rtab-map standalone, the settings of the above link work great even without depth calibration or image calibration. Odometry is very fast and accurate with IR.

However when I use RGB odometry things get messy. Everything is slower of course because of the data increase but loops are more difficult to happen and clouds get misaligned very easily.

Since I noticed that in the Azure Kinect Viewer v1.4.1 it is possible to capture at the same time in real time  both RGB and IR images, I was wondering if the same can be done in Rtab-map, ie: work with IR odometry but at the same time capture the RGB images. Then at post processing the IR clouds cloud be replaced with the RGB data but with the IR odometry that seems to me that is more accurate.

Does it make any sense ? Can this be done ? Any volunteers !?! ;P

Apart from this, are there any recommendations for rgb settings for Azure Kinect, or maybe a *.ini share of settings ?
My objective is to have the best possible spacial accuracy in the final cloud.

Thank you !

[matlabbe]

Hi,

If you are using ICP odometry, you could enable color (uncheck "Use IR for RGB image (may work better in dark areas)."). Odometry will still be computed with geometry, but you will have color at the same time. The downside is that you lose IR field of view (in general larger on those kind of sensors).

However, if you want to do vidual odometry with IR while recording RGB image for visualization purpose, it is not possible. This is something I wanted to explore last Christmas but I didn't have time to do. It would be  useful for D435 camera too, where IR visual odometry is a lot more accurate than using the color camera, but we also lack of color afterwards in the generated 3D map. It can be done with ROS, as odometry and rtambap nodes can use different topics. For standalone, we would have to find a way to send IR data to odometry thread, then color data + odometry to rtabmap thread.

Not sure which rtabmap version you tried, there was a fix in this commit to improve RGB visual odometry with K4A.

I also published in latest 0.20.8 new default parameters that help a lot with HD images for smoother odometry (in particular for K4A). With older version, look at the Release Notes and update the corresponding parameters in your ini file.

cheers,
Mathieu