Depth offsets calibration

[saikishor]

The values from the depth offsets vary from one calibration instance to another and the values totally doesn't make sense to me, as they are pretty bad. for instance:

z_offset_mm: 1143
z_scaling: 286.57693192836575

Surely, I cannot depend on these values, because as per the cfg in openni2, the max and min of z_offset_mm is -200 to 200, and z_scaling is from 0.5 to 1.5. Why is this happening?. Is there are standard procedure that need to be followed. The above experiment is done on a device similar to Xtion.

[ahoarau]

Strange, I remember this to be working fine. @JimmyDaSilva ?

[saikishor]

Thanks for the imminent response @ahoarau. I have calibrated the intrinsic parameters earlier both for the RGB and depth, I am not sure where the issue is. Can you please tell me, whether during the experimentation the aruco should be moved along one plane, or can be moved in 3D?. Thank you,

[JimmyDaSilva]

Hello @saikishor. This was an attempt to get better results with the old Asus Xtion sensor.
The code has been coded quickly and is not perfect at all.

If you really need to do this as well (which sensor are you using ?), then my advice would be to do as follow:
-1) cover the aruco code before starting
-2) hold the aruco flat with respect to the camera frame
-3) never get under the minimal pointcloud distance of your sensor ( 0.5m for the Xtion)

Explanation for people who might try to fix this code :
If 1) maybe the code can get stupid depth values at start
If 2) Not sure the depth value if read in the right pixel. I remember having a shift between the dot on the image showing the tag detection. So having the aruco flat will minimize this error
If 3), will you get a 0 or infinite value, or max values depending on if you use openni, openni2, or something to start the rgb-d sensor, and maybe I don't filter those out

Sorry but I haven't been coding on this for long. It sure does need to be coded properly, but I don't know when I will have the time to do it.

Feel free to ask questions, or send pull-requests anyhow.
Jimmy

[saikishor]

Ok I will try to take a look, I am also using xtion, but I wonder how you got good results. I am going to try with the guidelines you have provided me. When I keep it totally flat. I have something like this haha.

slope: nan
z0 nan
slope error nan

I will give a try with your guidelines, and update you.