- This topic has 3 replies, 2 voices, and was last updated 3 months, 3 weeks ago by
.
-
Posts
-
This is more of a request on help with modification of the original program.
Due to my design restrictions of the sandbox environment I am unable to place the kinect and the table close to each other. There is roughly a 2.5 meter distance between the table and the kinect. Due to this, the disk size remains insufficient as it cannot be detected fully, meaning I can’t complete the projector calibration phase.
I have checked and it appears that larger circular objects perform well (tried with a circle of ~10cm radius) but the program does not recognize these as valid disks. I would like to modify the code such that these are recognized. Any guidance with this would be appreciated.
I have checked calibrateProjector.cpp and it looks like valid disks are
I have read a few old posts and managed to find a solution to this.
In calibrateProjector.cpp, line 362, the radius can be manually set. You can also set what I assume is the error margin in line 363. I’ve modified these to suit my disk instead.
I have then run the make file again to compile these instead of doing gcc fiddling manually and it seems to have worked. I also don’t seem to need to do the previous calibration steps again.Unrelated, it looks like I did somethings wrong in the corner registration step (or I actually do need to run through the previous calibration steps again, we’ll see). Will attempts to fix that and then I think this will be ready!
Follow-up to this question: I the increased disk size has helped make the calibration a bit easier. There is still a problem when using it in a certain distance range but this has helped.
I now run into another issue with my overall calibration where some parts of my elevation map are perceived at the wrong height. I have uploaded images of the table to the following imgur extension in case the link causes issues: /a/sIk6kFd
I have measured the table and know that the top right (green plateau) section is indeed flat, there should be no issue of it being slightly slanted and being raised because of that. I believe the issue lies in this section being closer to the edge of the Kinect, which causes inaccuracy issues during calibration/detection.
As described previously, modifying the kinect is difficult in its current setup, so finding a software solution would be preferable. Additionally, upon visual inspection, the Kinect shows no signs of a tilt which may cause this inaccuracy to be only at the top right corner. If I had to move the Kinect, I am unsure what action would help this issue or how I could tell the setup was done correctly.
This is why I would prefer to solve this issue through software instead. Maybe add a slight negative bias for the upper corners of the Kinect data. How feasible would this be? I would like to hear some feedback on this, and if possible some pointers to where I may find the depth data I would need to modify in this case.
I’m sorry for my late reply, I did not see your Dec. 8 post until today.
Good work on the disc size change. And thank you for posting the images of your sandbox. For me, it’s easier to see issues than to interpret verbal descriptions.
Are you using a Kinect V1 or V2?
I believe the issue lies in this section being closer to the edge of the Kinect
This is plausible. I’ve seen pixel problems along one edge of a Kinect. The only solutions I know of are to replace the Kinect or shift/tilt the Kinect so that the bad pixels are outside the box.
upon visual inspection, the Kinect shows no signs of a tilt which may cause this inaccuracy
The physical orientation of the Kinect has only a small effect. The calibration procedures measure the sandbox base plane as read by the Kinect in its orientation during calibration. As long as the Kinect is not repositioned after calibration, things should work.
If I had to move the Kinect, I am unsure what action would help this issue or how I could tell the setup was done correctly.
If it were me, I’d do a quick-and-dirty prototype with a repositioned Kinect. I’d hang the Kinect from rods and clamps and re-calibrate, then see how the sandbox performs. Tweak the camera position and try again. I’d try to demonstrate (to myself) that it was worth the effort to make a permanent jig to re-mount the Kinect.
- You must be logged in to reply to this topic.