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.
Well, for those who are working with libfreenect on ARM or AARCH64 architectures (mainly Jetson TX1) because OpenNI and SensorKinect are problematic to build, I made some adjustments on OpenNI and SensorKinect sources to run with Aarch64 and avoid having to use libfreenect.
Links: OpenNI for TX1 and SensorKinect for TX1
