Machine vision targeting shapes

I want to have my webcam pointed at the screen and have it recognize shapes. Can you point me in the right direction

I dont have the software yet, I'm just asking questions about it.

I was challenged by a friend that I couldn't never become good enough to beat him in Battlefield4. Sadly this is true. But given my never give up personality, I decided if I could beat him with a crutch.

I want to have my webcam pointed at the screen and have it recognize shapes and have it move the cursor to that particular shape. After this is accomplished I want to customize a PS4 into a PC.

The PC will have a secondary more powerful and faster system so it can eavesdrop on the PS4 during FPS games like Battlefield 4

The PC will have a large SSD full of processed images from the game in a database. The enemy is easily identifiable by the red triangle about its head

With the correct type of machine learning and visual identifying processes the PC should be able to target the enemy by parsing the info on the screen. Allowing the me an auto aim feature (through CronusMAX)

With a CronusMAX and the correct python and machine learning programs it should be able to learn how to play the games on the PS4. I would never sell this machine It will be too costly to build and code. It would just be my personal crutch If i could make it happen.

Answer 1

[Image Analyst]

The Computer Vision System Toolbox has lots of demos in it on tracking people and vehicles. You can try some of those. This will be a long and complex problem that may take you weeks or years to almost get right for the wide variety of things that will move in your scenes.

[TwentyG67]

what type of processor or system will i need to run this on that large of amount of data?

[Image Analyst]

The toolbox can normally run at near real time (10-30 frames per second), depending on what kind of analysis needs to be done on each frame. If the computations are extensive, then of course it will be slower. I'd try it on your main computer first. If it's too slow, like less than a frame a second or whatever, then you can try getting a faster computer.

[TwentyG67]

If it's needed to track 2 targets and the computations are extensive will the CPU need more raw power or more threads

[Image Analyst]

Depends on how complicated the thing you're looking for is. If it's just anything that moves, that's pretty easy and fast. If you have tons of moving things and you need to figure out what moving blob is a jet and what is a soldier and what is a tank, then that can get pretty tough and require lots of computation time.

[TwentyG67]

computation time = what? lets cut to the chase, if it requires a lot of computation time are we talking more memory, threads or raw power?

Intel Core i7-6950X 25M Broadwell-E 10-Core 3.0 GHz LGA 2011-v3 140W BX80671I76950X Desktop Processor.

10 Cores / 20 Threads

LGA 2011-v3

4 Channels DDR4 2400; one DIMM per Channel

140W TDP

Up to 40 Lanes PCIe 3.0 (2x16+1x8)

25MB of L2 cache

would that do it or is it overkill?

[Image Analyst]

I would think that should be sufficient.

Answer 2

[Image Analyst]

See my attached shape recognition demo.

[TwentyG67]

I edited my question please re-read

[Image Analyst]

Note: My above answer was for a totally different question. The poster totally changed the question.

[TwentyG67]

To be fair the original question is still the first sentence. I just realized that it needed clarification which changed the answer. Thanks for your reply