How to recognize specific objects

asked 2017-09-03 04:19:58 -0500

holdman gravatar image

I have been making application which recognize car light symbols on Android in Java using OpenCV. I take the images from phone's camera. I use OpenCV 3.2.0 version. What I have done so far:

  1. Filter out the colors in an image
  2. Reduce the noise
  3. Detect an object by size
  4. Sort out the object by color(the object will be compared with the same object's color)

Detected object:

Detected object

Colors filtered out:

Colors filtered out

What would be good option to recognize those (binary images) objects? I took a look at HOG, HAAR and k-NN algorithms. But I struggle to get informations about how to train them. Or do I miss something? I do on windows 10 everything. Let me know your opinions.

edit retag flag offensive close merge delete


i'd try much simpler means first, like matchContours(). opencv also has a shape module for this

berak gravatar imageberak ( 2017-09-04 01:30:38 -0500 )edit

But wouldn't it be realible enough to recognize the objects? And how is it going to be with speed?

holdman gravatar imageholdman ( 2017-09-12 09:41:07 -0500 )edit