Ask Your Question

How to use Train cascade?

asked 2016-02-03 19:53:27 -0600

bob409 gravatar image

I have to detect hand in a video file hence I think that train cascade could be used to do that. I have already taken some pictures for positive and negative images. What are the next steps to do? I am using opencv 3.0 and c++. If you could give me a guideline on what i should do, it would be better.

edit retag flag offensive close merge delete


i'm having some doubts, if cascade training is the best approach for hands ( due to rotation/pose issues ) but have a look at the tutorial

berak gravatar imageberak ( 2016-02-04 00:16:54 -0600 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2016-02-04 04:56:06 -0600

updated 2016-02-04 04:57:08 -0600

I can confirm that cascade classifiers are far from useful in this case, since the tend to model rigid objects, unless you are using a template combination of multiple cascade classifiers all responding to a specific orientation.

More frequently used setups for hand detection:

  1. Color based segmentation. The most basic accepted rule for skin segmentation is the RmG rule --> 20 < R - G < 80
  2. There is the skin classifier based on Naive bayes classifier, which is built on the dbskin dataset. This could be trained in OpenCV also.
  3. Finally the Piotr Dollar ICF/ACF framework showed that tranferring the image to the LUV color space allows for easy segmentation in the U channel.
  4. There is also a deformable hand model available using the DPM framework cited by Mittal.
edit flag offensive delete link more

Question Tools

1 follower


Asked: 2016-02-03 19:53:27 -0600

Seen: 194 times

Last updated: Feb 04 '16