flammxy's profile - activity

I want to training data and use HOG algorithm to detect pedestrian. Now I can use defaultHog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector()); in opencv to detection, but the result is not very good to my testing video. So I want to do training use my database.

I have prepared 1000+ positive sample, and 1000+ negative samples. They are cropped to size 50 * 100, and I have do the list file.

And I have read some tutorials on the internet, they all so complex, sometimes abstruse. Most of them are analyze the source code and the algorithm of HOG. But with only less examples and simple analysis.

Some instruction show that libsvm\windows\svm-train.exe .can be used to training, Can anyone gives an examples according to 1000+ 50*100 positive samples?
For example, like haartraing, we can do it from opencv, like haartraining.exe –a –b with some parameters, and get a *.xml as a result which will be used to people detection?

Or is there any other method to training, and detection? I prefer to know how to use it and the detail procedures. As the detail algorithm, it is not important to me. I just want to implement it.

If anyone know about it, please give me some tips. Thanks

error infomation: OpenCV Error: Assertion failed (elements_read == 1) in unknown function, file .. ....\opencv\apps\haartraining\cvhaartraining.cpp, line 1859

I am doing pedestrain detection using haartraining. My positive samples are 900 samles, each is 40x80 , with information ( 1 0 0 40 80 ) and 500 negetive samples, each is 300x200

I have check the parameters, they are all ok. But when doing opencv_haartraining.exe it always shows this error. Could anyone help me?

Thank you

For people detection. where can I get the location and size of rectangle of pedestrian of PETS dataset?

Thanks

I am now doing tracking object frame by frame using mean shift.

But I found that mean shift can tracking deep colour object but can not tracking light colour object.

Maybe the back-projection cause this problem.

Does anyone know some other simple and fast tracking methods that can track object frame by frame?

Thank you.

Does anyone know how to tracking white object by mean-shift?

I try to tracking object by mean-shift, but find that if the target object is dark or deep colour, it works good.

But if the object is white or light colour, it does not work. It seems like back-projection cause this problem.

How to solve this problem? I do not know how to solve it.

Does anyone have some experience about it?

Thank you.

Most examples use hue channel to track, but if the target object is white, the hue value is 0. And the result rectangle will shift far away from the object? How to solve it?

Thank you.

Does anyone knows how to do it? Help me please. Thank you

I tried it.

But find the result of ch[3]={2,1,0} is better than ch[3]={0,1,2}, and the result of ch[3]={2,1,0} is still not good. Do you know why it is?

Thank you.

I am doing it as a part of mean-shift tracking.

calcBackProject( const Mat* images, int nimages, const int* channels, const SparseMat& hist, OutputArray backProject, const float** ranges, double scale=1, bool uniform=true );

The 3rd parameter is: int* channels.

if do it in 1 channel, I know it is define as: int ch[1]={0}, now I want to do in 3 channels (input a RGB 3 channel color image), how should I define it?

int ch[3]={2} ? or int ch[3]={0,1,2}? or int ch[3]={2,1,0}? or int ch[3]={2,2,2}?

Does anyone know it and Could you give me some tips? Thank you.

Usually examples and codes is convert RGB to HSV and split hue channel. I dont know the reason. Can anyone tell me the reason or is there any detail theory or examples about it?

Thank you

Hi, thanks for answering my question. The problem I meet is that the results are not good especially when the majority part of the objects I tracking are white, lose the target even 1 frame late. For example the people wear white clothes.

Or when the color of majority part of people's body are not high contrast to the background. So from the test result, it does not work good.

I have found some examples and codes about mean shift. But they all using hue channel of HSV to tracking. The first step is to convert RGB to HSV. But I have the problem that the results are not good especially when the majority part of the objects I tracking are white, lose the target even 1 frame late.

Does anyone know how to mean shift tracking by RGB 3 channels? Or are there any samples or codes about it?

Thank you.

Yes, I build it by using cmake, and than find createsamples.exe.

I am studying the haartraining. From some papers, the author says opencv has it's own createsamples program. But I can not find createsamples.exe or any other createsamples files in the opencv folder. Does anyone can tell me some information about it?

Thanks a lot.

I have just learned blob tracking and HOG algorithm about car detection. Could anyone gives me some tips of any new and good algorithms? Thanks

Thank you. But sorry I do not know how to use Latent SVM.

Are there any new and good algorithm about pedestrian detection,people counting?

And if I use HOG descriptor to detect pedestrian, are there any other tips to improve the result? Because I find my result is not very good.

Thanks.