Friday, 15 March 2013

opencv - people detection with haar cascade -


I am working on a project in my school to find out how many students are in the class. Like this picture class

I identify people to identify faces I'm trying to use Higher Cascade in opencv, but the result is very bad: find out After

I took thousands of pictures in the class and cropped the picture manually with about 4000 positive samples and 12,000 negative samples in it. I was wondering what I did? When I crop the image, should I just harvest this kind of crop? head or body with this kind? with body

I think I have adequate training samples, and I With this precise process, this post: What should work or should I use a different algorithm such as HOG or SVM would be very good for me for any suggestions, I have been stuck in this for months and No clue, thank you very much!

Haier is better for human face. With the SVM the hog is classic for human identities and about them There are many sources and blogs, it is not difficult to train a classifier, for your visualization, I think that "head and shoulder" is better than "head alone" but your multi-visual sample enhances the difficulty. The camera will be better if you are always more false positive If there are Mac alarms, then add more difficult moist samples. This paper can help:


Posted by at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)