Upper Body Detection w/ Kinect v2
Opened this issue · 4 comments
Is there any specific configuration to upper body detection work with Kinect v2?
I was able to have this working with depth map from stereo but with an high accurate sensor as Kinect is.
On my point view the two main causes could be:
- Camara Parameters (K matrix)
- Ground Plane estimation (in my case is fixed [0,-1,-0] at 1.5m from the floor .
But after confirming these two causes no better results were found. Instead it has lot of false positive in some high cabinets that we find in the sequences.
Is there anyone who could be able to put this detector working with K vs2?
Sadly, I am currently not able to do this. Maybe @lucasb-eyer has some experience or at least a helpful tip how to achieve this?!
I think the best way to debug this is to dig into the code in order to
first visualize the regions of the image that are actually considered based
on the ground plane projection and then the actual reactions based on the
upper body template. I'm not familiar with the details of the code, but I
wouldn't be surprised if certain assumptions were explicitly optimized for
the xtion sensor. As the resolution of the Kinect v2 is quite different and
it also has different characteristics, it could be that you need to adjust
stuff to fit your needs, but for that you will need to understand what is
going on and look at which point in the pipeline it fails.
On Mon, Nov 2, 2015 at 11:54 AM, Christian Dondrup <notifications@github.com
wrote:
Sadly, I am currently not able to do this. Maybe @lucasb-eyer
https://github.com/lucasb-eyer has some experience or at least a
helpful tip how to achieve this?!—
Reply to this email directly or view it on GitHub
#183 (comment)
.
I don't have anything more useful than what @Pandoro said above, sorry. I also don't currently have the time (and incentive) to make it work on Kinect v2.
Cross-linking to spencer-project/spencer_people_tracking#4 which asked the same thing.