i have had past experience with the pru’s of beaglebone and are found to be of great use.
i have seen that till date no single board computer is well enough to provide stereo vision using two cameras fast enough for real time application.
it has wide application in robotics for stereo vision depth sensing
presently we have kinect camera as the best option but in not open source and cost effective
intergrating stereo camera with ros will open room for a wide audience
very much in need for MAVs (micro aerial vehicle) with almost real time capability
PRUs are best suited for this task and no other board can do such fast processing while managing to offload the work to PRUs and creating space for other softwares like openCV
i applied last year too for this project but was not selected.
how useful does the project seems? can i propose this for gsoc 2017?
Did I miss where you estimated the number of cycles to fetch the frames from the PRU, put the processed data back and do the processing itself? Some sort of rough calculation of the number of mathematical calculations is all I’m seeking, plus known bandwidth/latency to manually fetching the frames (or a realistic plan to utilize DMA).
Anyway, I like the project idea a lot, but not sure I could mentor it personally.
This seem a bit large of a project for GSoC. Prehaps break it up into a few
projects and slate one of them for GSoC:
- PRU camera interface. I can help with that. For stereo cameras, this can get
tricky. So a prereq. is to have some kind of timing estimate (based on
resolution and frame rate).
- Stereo processing. Prehaps via the GPU. You will need to express things in
terms of OpenGL ES2 on most of the boards.
- A third project that builds upon this (outside of GSoC) to do what you are
describing.
I have listed these (in a more general form) as 2 ideas on the GSoC page. The
PRU camera interface may might be further divisible into 2 sub projects:
a) The actual data acqusition itself.
b) A V4L2 interface.
One thing that would be needed on the student side is a realistic timing
estimate for the scope of work undertaken.
I suspect this will have to be X15 specific.
Btw - there is also experimental stereo image processing in libpcl so it isn't
just OpenCV.