Closed goldhoorn closed 9 years ago
Well, the depth deconvolution is extremely calculation-intensive. To get any near-realtime framerate at all, you need to use shaders or OpenCL for processing.
Sorry, forgot to mention that there is a CPUDepthPacketPipeline, but even a big Core i7 won't be sufficient to get realtime processing.
I just figured out by adding your driver to rock (www.rock-robotics.org) that it has several dependencies to GL stuff.
Would it not make more sense to create a "clean" only driver, instead mixing this with visualization stuff? The real-driver seems to be in examples/ and there the "libfreenect2" target from the CMakeLists. Do you have any reasons for this structure? I would like to create a "clean" only driver. The folder structure could be like:
src/ include/ test/
Where in test Protonoect goes. I would like also like to change the DefaultPacketPipeline to NOT use GL stuff. And create a GLPacketPipeline instead. The GLPacketPipeline headers could then be activated or deactivated by Cmake. This would reduce the amount of dependencies libfreenect has.
Before i start with this i would like to have the opinion from the responsible "core" developers.
Best, Matthias