Hi Folks,
Our team used a Raspberry Pi for vision processing last year. The student ended up with code that had pretty poor performance; about 1 frame per second. In the off season, I and a few students spent some time working up code to explore the performance and try to optimize the code.
We've got a body of code I thought I'd share. I'm afraid it's not complete, and it's really only suitable for someone with a strong C/C++ background.
The code is all on GitHub here:
https://github.com/Team-2823/vision2
It has one novel useful feature - I rewrote the OpenCV capture function in C, with a built in color filter capability, so in theory we can get a frame in more rapidly than any other way. (Although frame capture turns out not to take that much of the time; it's things like canny and finding contours that are expensive). It does have a lot of other features; timing of functions, configurable functions to perform (e.g. blur, canny, contours, fast, and so on), and tuning for color and camera filters.
But it's more of a toolbox; it's not at all a ready made set of vision code, and really suitable to someone who feels pretty comfortable in C and C++.
At any rate, I figured I'd share instead of just sitting here refreshing the kick off countdown timer... <grin>.
Cheers,
Jeremy