Logo detection OpenCV
I am doing project about logos recognition in input video. I have tried many ways but all of them give me wrong solution. I have used: -SURF/SIFT descriptors/detectors but the number of key points is too small and this fail - any other descriptors/detectors like FAST/ORB/BRIEF but with the same result - matching template, but this is not scale, perspective, angle resistant - i have read about HAAR classifier but this is not good for my project because the set of logos to recognise will be changing and in Haar i must to train with positive and negative images
Is there any other possibilities to solve this problem? The example logo http://pl.tinypic.com/r/2r3jryx/8
A simple answer: bag of words object recognition.
Will bag of words work in app where user can add logos and select videos, then app detect the logos?
Yes it will, you only need room to store your vocabulary.
How to implement this? any source code are available? i can't find any. I have a folder with images(logos) and video. I must implement app which can recognize the logos on every frame from video. I would be grateful for help.
This link is all you need using C++ and JNI to run it on Android!
Thanks, but i know how to use OpenCV and how to run a sample. I would like to know how to implement BOW for my case.
By going to the equivalent C++ example which is linked in the documentation, top of the page? Please do put in some effort ...
@StevenPuttemans , btw, i think we need a better code example here. the one you're refering to spends 90% of ~2500 lines of code on parsing extremely useless pascal/voc datastructures
Aha :D Feel free to make a new one! Also, @GilLevi has a site with quite some nice guide on BOW