Cat Detection Using Haar Training

Question

I would like to detect deformable objects (a CAT in my case) , using HAAR- cascades . I ran the test on 2000 positives and 3000 negative images upto 20 stages . I don't see anything expected happening. I am sure it might be the problem of my training data set. My positive data set has the cat images in it . In different shapes and postures of the same cat . I wonder if Haar would be able to recognize the differences in the positive data set , and still recognize them as cats.

Need help , Is there any other alternative method that I could use ??

Answer 1

First off, you haven't listed what you are using to train your haar classifiers. If you are not using opencv_traincascades this is the place to start.

Secondly detecting a cat is a very difficult job, there are a lot of undefiend edges, rotational and spatial variances. using this train of thought, answer these questions to yourself:

• What does your positive dataset look like?
• Is this similar to what you are showing through the input feed
• What is the false alarm rate of your classifier when you have trained it?
• What are you including in your background set (more images that look like cats that arn't cats are better etc)

I would suggest posting your output into the question from the classifying stages in order for us to get a better view of what is happening. Take a look at these links for more help.

Maybe try with pens, or logos first, follow the openCV tutorials and understand the basics? (sorry if I have assumed you don't know the basics if you do)

G./