First time here? Check out the FAQ!

Ask Your Question
0

Cascade training(Haar training) - I need help!

asked Jun 14 '13

Hi i'm currently working on a AR Drone project running on ROS+Ubuntu. I'm using the cascade for the Drone camera to track a logo/image and follow it. However, it does not track the image well when i use haartraining,it will track other things easily instead of the logo. I'm not sure on what parameters i should use. I'm a beginner at this, i followed this guide http://note.sonots.com/SciSoftware/haartraining.html .

This is what i typed to create the cascade, opencv_createsamples -img -bg bg.txt -vec out.vec -w 24 -h 24 opencv_traincascade -data out -vec out.vec -bg bg.txt -numPos 950 -numNeg 400 -numStages 15 -w 24 -h 24

Are my parameters correct? Should i add on more? My cascade does not track the logo/image well. I need it to track the logo well. I used on image of the logo to create the samples.

Can anyone help me? I would really appreciate the help, i've been trying this out for 1 week, trial and errors. But i still can't get the drone to track the image well!

Preview: (hide)

1 answer

Sort by » oldest newest most voted
1

answered Jun 14 '13

Immi gravatar image

updated Jun 14 '13

I am also not so experienced in Cascade training, but I think you should use more than 15 stages. When I tried it, I got with 30 stages good results. And I'd take more negative images.

Did you try feature tracking for your logo? I think this could be also a solution. http://docs.opencv.org/doc/tutorials/features2d/table_of_content_features2d/table_of_content_features2d.html#table-of-content-feature2d

Preview: (hide)

Comments

I agree with Immi. Here is a good database for negative images if you need:

http://tutorial-haartraining.googlecode.com/svn/trunk/data/negatives

-Vinay

vinayverma gravatar imagevinayverma (Jun 14 '13)edit

Question Tools

Stats

Asked: Jun 14 '13

Seen: 570 times

Last updated: Jun 14 '13