Comments on: OpenCV Haartraining: Detect objects using Haar-like features

By: David

David — Sun, 17 Jul 2011 16:11:45 +0000

Hello,
Iam new to opencv. My english is poor.@@
I have same problem as Srinivas A said on 15-04-2010

In haartraining some times I get Error: bad flag (parameter or structure field) () in function cvReleaseMat, cxarray.cpp

What does it mean?Why it happened?

By: Viet Anh

Viet Anh — Fri, 24 Jun 2011 02:28:10 +0000

Hi everyone,
I have successfully trained a face detector using Haar feature. Does anybody know how to use LBP feature instead of Haar ? (with openCV, of course). Thanks in advance

By: Luis

Luis — Thu, 02 Jun 2011 17:32:20 +0000

Hi Andrey:

I’m trying to understand how this can be implemented but I’m not sure how can I do it. I would really appreciate if you could see what I understand and correct me if I’m wrong.

I understand I have to create a predefined set of features lets say 1,000 of them and then, try one classifier per feature so I start selecting which features are the best following ADaboost method.

What I’m not sure is how to apply the feature on the integral image, I mean, once I have integral image I have to divide it in subwindows right?.

If the base would be 20×20. should I have no overlapping windows or should I have as many subwindows as possible??.

For example, my first subwindow would be at coords (1,1,20,20)… then, the second one should be at (2,1,21,20) or at (21,20,40,20) ???

And the last question, once I have my subwindows, how do I apply my features? Should the feature fit into the whole subwindow or it can be smaller? In case it can be smaller, Should I pass it position by position all over the subwindow??

I’m sorry but the document of Viola doesn’t seem too clear to me.

Thanks in advanced

By: Hani

Hani — Mon, 09 May 2011 08:05:47 +0000

Hi, Hannah,
I have the same problem with you. Have you solve the problem later !!! If you have any solution please help me! Thanks.

By: michaela

michaela — Thu, 05 May 2011 16:23:27 +0000

Hi,

I read that the variance normalization is suggested by Viola and Jones as a mean of reducing that effect of different lighting conditions.
So, for to realize an implementation of Viola and Jones, the threshold value given in the .xml must be multiplied for the variance….

I’m not sure….Someone can explain it better?

By: michaela

michaela — Thu, 05 May 2011 16:22:55 +0000

Hi Edward,

I’ve the same doubt…
I read that the variance normalization is suggested by Viola and Jones as a mean of reducing that effect of different lighting conditions.
So, for to realize an implementation of Viola and Jones, the threshold value given in the .xml must be multiplied for the variance….

I’m not sure….Someone can explain it better?

By: Juliano

Juliano — Sat, 09 Apr 2011 23:58:49 +0000

Hi, the positive images for haartraining can be a large image with the object inside? or positive images only can have the object?

By: sid

sid — Sun, 27 Mar 2011 18:14:33 +0000

Any answer to Edward Li’s doubt? Anyone please.

By: Sid

Sid — Sun, 27 Mar 2011 18:13:09 +0000

Hey, does anyone knows how to use the xml file threshold values exactly. I am making a Hardware implementation of Viola jones and its very important to know the exact way to use the threshold value.

What i perceive is — Calculate variance – calculate Standard deviation from it – and multiply it with the Threshold and then compare it with the normal feature value

Is it correct or there is something more.. some normalization factor of some sort etc..

sid.

By: Aleksan

Aleksan — Mon, 24 Jan 2011 07:41:04 +0000

Hi all.
Dose anybody build OpenCV with TBB support ?
Dose in work ?
I try but with out any result. I build OpenCV with ЕИИ support, it linked to tbb.dll, but when running OpenCV use only one core and work only with one thread.
Is in possible use multicore when training haar files.