I've been using OpenNI+PrimeSense+NiTE with OpenCV on my project to segment objects according to their distances. However I meant to deploy it in a NVIDIA Jetson TX1 board and it couldn't manage to compile OpenNI+PrimeSense+NiTE with OpenCV on it. I endded up with libfreenect. However the depth map provided by libfreenect is very, very wrong. I'll share some examples.
Here is the working depth map of OpenNI: OpenNI Depth Map
The libfreenect wrong depth map is here: Libfreenect Depth Map
I based my libfreenect code on the default C++ wrapper at OpenKinect website.
Can someone help me here? Thank you so much.