Jetson TK1 deployment #4
Description
As suggested by @wiedemeyer, here are some deployment notes on Jetson TK1 I sent him earlier. I will follow up here.
Jetson TK1 offers 300 Gflops at 10W with USB 3.0 and GbE. It seems perfectly capable of handling Kinect2 data while consumes little power. But Nvidia provides no OpenCL support for Jetson TK1 so this project will need some CUDA code (I might write some).
Compiling this without OpenCL will requires some changes to the code. Besides some tweaking of CMakeLists.txt, @wiedemeyer suggested to
comment out "#define CV_OCL" in line 31 of the kinect2_bridge and it will not use any opencv opencl functionality at all. When the depth registration is initialized in line 143 of the kinect2_bridge you can choose CPU instead of OPENCL. And in libfreenect2 change the depth processor to the CPU or OpenGL one.
However Jetson TK1 CPU is weak. I don't think it can handle depth registration by CPU alone.
libfreenect2 compiled but didn't get any data from Kinect2. Protonect got stuck waiting for the first frame, after good initialization. dmesg had some warnings (add later) about tegra-xhci. This is hard to debug. It's possible this new device and new driver just don't work with Kinect2. I'll get a NEC uDP720202 and see if this one works. And then test this project.