Hi, I’m working on a hobby project, making a tablet top interface for multiple users, which requires multiple hand tracking.
A Kinect is mounted above the table, capturing images of top view of the table.
I’m making this using Kinect + openFrameworks + OpenNI (ofxOpenNI). I wondered if it’s possible to track hands using OpenNI/NITE’s hand generator, but it seemingly needs body image, not only hand + arm image.
I drew a pic for better understanding what I’m saying here, please look at the attached image.
There are multiple hands moving freely, interacting with the table-top interface.
A hand can be on the top of another, which makes the situation complicated.
When this happens, and if the image of hands (with the background subtracted) is converted into a binary image, it appears as a blob occlusion.
This, occlusion occurrence possibility, is the difficult point for me.
Without occlusions, I think I can do what I want to do. Just subtracting background by depth threshold first, and making the image into a binary image, then each blob represents hand+arm, then just tracking those blobs. That’s it.
I read lots of papers on occlusion handling but they seem difficult and too heavy to implement for me.
Does anybody help me please?