Object Detection project with OpenCV

Question

I am trying to make an Object Recognition software for my final year project using OpenCV. After studying quite a lot about this field, I found out the plan of action should be this:

Extract features from a lot of images.
Create a training dataset from the extracted features.
Label the dataset.
Make the machine learn from the dataset.
Test the model.

I started from a single image. I was able to extract features from the image using SurfFeatureDetector class of OpenCV (simple_matcher.cpp program given in the samples). I saved the detected KeyPoints in an XML File using the following code:

SurfFeatureDetector detector(400);
vector<KeyPoint> keypoints1, keypoints2;
detector.detect(img1, keypoints1);
detector.detect(img2, keypoints2);
FileStorage fs("test.xml", FileStorage::WRITE);
write(fs, "data", keypoints1);

Now I am stuck at this point. I am not able to understand how do I create the training dataset from these features? And what should be my next step? Or, is my plan of action correct?

Thanks in advance.

Format and content of a training dataset depends on what you want to recognize. Please, specify your goal more clear. — old-ufo
– old-ufo, Commented Nov 20, 2013 at 8:44
I want to detect objects in an image, which may be multiple in number. So, i want to create the training set with, for example, 100 images with a ball, 100 images with a tree, and so on. — R4chi7
– R4chi7, Commented Nov 20, 2013 at 8:46

Darshan · Accepted Answer · 2013-11-20 09:56:09Z

4

you would probably need to do these following steps.

Find out and zero in on a feature detection and extraction algorithm you want to use(SIFT,SURF,ORB, there are few more).
Detect and extract features. you should be storing descriptors extracted from a descriptor extractor in your xml files.
add those descriptors to a trainer such as bowtrainer to create a dictionary.
Train your classifier

You can find lot of examples here and you can watch this video if you want to realize something similar in your project.

answered Nov 20, 2013 at 9:56

Darshan

1,01810 silver badges19 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

R4chi7 Over a year ago

i used the BOWTrainer to create a vocabulary. But now, I cannot understand what to do next with the vocabulary. Please help.

Darshan Over a year ago

Now,start training your classifier. You can look at thisexample

old-ufo · Accepted Answer · 2013-11-20 09:19:43Z

3

First, it would be easier to use popular datasets, for example, from here. It allows you to not spent time on images and gives a possibility to compare your results with state-of-the-art.

Second, it is not easy task. I recommend you to read papers related to the PASCAL VOC.

If you want to do the first try, read about bag-of-words model, for example it is pretty short and nice explanation what to do in the blog of the SO user @gillevi.

answered Nov 20, 2013 at 9:19

old-ufo

2,8602 gold badges32 silver badges43 bronze badges

Collectives™ on Stack Overflow

Object Detection project with OpenCV

2 Answers 2

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related