In an external testbed I have I was able to test my vision code and make some minor patches (not shown in github at this time). It appears that I can find objects and "follow" them but that last part is rather untested. The better news is at 2.33GHz I get about 6.3fps post processing so at 1GHz I should get at least 2.5fps. Some quick number crunching at a walking pace shows the car will move no more than about 3' in a single frame which while a lot is better than expected seeing as I'm not using a computer.
Fact is though, that pushing all this onto the BBB is... not easy. openCV can technically be cross compiled but that is not very easy to do seeing as the Intel processor on my computer won't recognize any ARM files. So I can either look at openCV code on an Intel install, or I can upload some ARM code. I'm not really sure how to get both it seems that every time I try to something breaks. I would send links to everyone but all I've successfully done so far is completly break my make files for my openCV installs.
Next strategy. Well maybe if I copy all of the include and libs from my BBB onto my Ubuntu machine it will maybe have all the ARM files it will need including the openCV ones that the BBB came with.
So the coming strategy is as follows. maybe if I copy over the project onto the BBB and than build it from src on the BBB all of the needed dependencies will be there. This is really not ideal though. Seeing as how my Ubuntu machine barely understands openCV after copying it from the BBB I can never compile my project on eclipse losing A LOT of the debug potential it offers. I'm going to continue hunting down ideas on how to compile from the Ubuntu machine but I'm running out of ideas. The only frustrating this is that derek molloy clearly did it somehow but I can not find him explaining how he installed openCV at any point.
Update everyone next time something important happens ideally this problem will be resolved by than.
No comments:
Post a Comment