Using a coprocessor for vision

[Greg McKaskle]

The Jetson board is relatively new, but some teams have incorporated other processors and placed their vision code on it. Searching CD for coprocessor will turn up half dozen different approaches teams have used. It will also bring up some of the issues they have to deal with in order to make their solution effective -- power, communications, and programming among them.

Computer vision is a hard problem, and adding CPU to it doesn't trivialize it. It will open up the door to using higher resolutions, higher framerate, and additional algorithms, but code that doesn't work on the cRIO probably doesn't work on the coprocessor either.

To review the options for vision:
The roboRIO will be about 4x faster than the cRIO.
The dashboard laptop is far faster than either of these, but you must deal with your limited bandwidth and communications delays.
An onboard coprocessor gives you a dedicated piece of HW to integrate with your camera, but there are challenges in getting it to work and keeping it working.

I'm not trying to talk you out of using a coprocessor, but I will also suggest that you look at your previous vision solution and determine why it failed half the time. Getting better at debugging vision problems is key to knowing how to improve it by whatever means.

Greg McKaskle