Clever Geek Handbook

Classification of OpenCV cascades with characteristic oriented gradient histogram (HOG) type

I'm trying to use an OpenCV cascade classifier based on an oriented object histogram (HOG) object type - for example, the document Fast Human Detection Using Oriented Gradient Histogram Cascade.

Searching the internet I found that the OpenCV Cascade Classificationator only supports the HAAR / LBP ( OpenCV Cascade Classification ) function type .

  • Is there a way to use HOG with OpenCV cascading classifier? What do you suggest?
  • Is there a patch or other library I can use?

Thanks in advance!

EDIT 1

I saved my search when I finally found in android-opencv that the Cascade Classifier has a connector that allows it to work with HOG features. But I don't know if this works ...

Link: http://code.opencv.org/projects/opencv/repository/revisions/6853

EDIT 2

I have not tested the fork above because my problem has changed. But I found an interesting link that might be very helpful in the future (when I come back to this issue).

This page contains the source code for Oriented Gradient Histograms for Human Detection. Also more information. http://pascal.inrialpes.fr/soft/olt/

+3


source to share


3 answers


It is now available in non-python code as well. opencv_traincascade in 2.4.3 has a HOG featuretype (which I haven't tried):



 [-featureType <{HAAR(default), LBP, HOG}>]
+1


source


If you are using OpenCV-Python you have the option to use some additional libraries like scikits.image which have a histogram oriented inline gradients.



I had to solve this same problem a few months ago and have documented most of the work (including very simple Python implementations for HoG and GPU HoG implementations using PyCUDA) on this project page . The code is there. The GPU code should be easily modified for use in C ++.

+3


source


Yes, you can use cv::CascadeClassifier

with functions HOG

. To do this, simply download it with the help hogcascade_pedestrians.xml

that you can find in opencv_src-dir/data/hogcascades

.

The classifier is faster and its results are much better when trained with the aid hogcascade

compared to haarcascade

...

+1


source






More articles:


All Articles