How useful could the Arduino Yun be in computer vision in a robotics project?

Question

I've recently ordered an Arduino Yun and several other components to begin on a small robotics project (based on an old 4x4 RC Truck). Although not mandatory for what I'd like to do, I was wondering how feasible it would be to mount a small camera or two and write some software for the Linux portion of the Yun to do basic image/video analysis such as face detection and computer vision based path finding?

Mostly this is a question about the processing power capabilities of the Yun. Has anyone had any luck doing simple image analysis/recognition on this processor?

(Also, if anyone with 150+ rep could edit this to create and add the "computer-vision" tag that would be great)

Answer 1

It probably has enough memory and speed to run basic object recognition software, but I there's a better solution, and that's to do your image capture and your object recognition with another device, and use the Arduino to act on the incoming data in some way. That would give you enough juice to do object tracking or searching. I plan to do this for my copter gimbal hopefully to be able to track objects on the ground, like criminals running from the cops for example.

This product will work great for that. Since you need a camera anyway, I think this is a pretty good deal... http://petapixel.com/2013/09/01/pixy-low-cost-camera-recognizes-follows-objects-color/

Can be purchased here: http://www.adafruit.com/products/1906

AdaFruit has tutorials to help you get started. (Computer-vision doesn't exist as a tag right now and I can't create it, but I tried, sorry)

Answer 2

For this application the Arduino YUN is the worst possible solution because

• Low RAM
• Terrible Processor
• No storage (you can add a microSD card though)
• High Cost

The raspberry Pi 2, UDOO, RADXA, Ordio, pretty much anything with a quad core processor and 1GB of RAM will do a decent job.

Face detection is RAM intensive, so you will need to worry about it lagging.

Alternatively you could also have the image processing part be done elsewhere like on a server (on a local network this could be your laptop) or something.

Personally,I would never recommend the Arduino YUN because it just simply isnt worth paying the 75$ for when for about the same price you could have a arduino pro micro and a raspberry pi (with a WiFi adapter, mircoSD card etc.) which is capable of doing probably way more than you actually need to.

Answer 3

Maybe you can consider adding Raspberry Pi board to run OpenCV which is well supported by hundreds of books and tutorials. See 20 tutorial from initial setting up to running typical functions in http://www.stemapks.com/opencv.html

Also, http://opencv-srf.blogspot.hk/

Then, there are many examples / tutorial / book on linking Arduino to RPi.

On the other hand, from web demo, Pixy CMUcam5 seems very good as a semi-ready-made solution. Watch the demo and see if it does what you want. If yes, then, it is easy solution. OpenCV is a library, it can do many tricks but need learning.

Hope this helps

Answer 4

I am guessing not very useful. Image recognition is quite CPU-intensive, and Arduinos are slow, memory-starved, 8-bit processors.

The Unix side of the Yun runs at 400 MHz, which is quite a bit faster than the Arduino part, but still quite slow compared to a mobile phone or desktop computer. And 64MB of RAM isn't enough to hold even a single HD image, never mind the data you would need to have in RAM to do image analysis. You'd probably have to limit yourself to 640x480, and the slow processor and lack of enough flash for large code libraries would be a problem. (You've only got 16 MB of flash memory on the LINUX processor.) I don't know if you can run programs from an SD card or not. That might solve your code storage problems, but you're still going to struggle with 64 MB of RAM.