OpenCV - Haar Training Resources
There were lot of good feed back for my OpenCV articles (Article1, Article 2). There were lot of inquires regarding Haar Training and how to do it. Since I was bit busy with my work at hSenind I couldn't put any replies for that. Finally I managed to find some time to put together few URL and important tools where you can use to create your own classifier XML.
If you are interested in OpenCV vision library you should join to OpenCV yahoo group. It is very active and has lot of resources as well. Following are some good links I used when I was learning on how to do Haar Training with OpenCV.
- http://note.sonots.com/?cmd=htmlinsert&page=haartraining
- http://opencv.willowgarage.com/wiki/FaceDetection
- http://note.sonots.com/SciSoftware/haartraining.html
- https://webeng.cs.ait.ac.th/cvwiki/opencv:tutorial:haartraining
- http://64.233.169.104/search?q=cache:PMA3ghxtCbwJ:rogerioferis.com/ClassMarch10/CreatingBoostedClassifiers.doc+HaarTraining+%2B+OpenCV&hl=en&ct=clnk&cd=48
Most important thing to consider when doing Haar training is to have good positive and negative image set. We have created a small utility modifying sample program provide with OpenCV to create positive samples. You can download the source here. Using that you can create positive samples to detect some object using a video. You can move frame by frame in the video using 'Space Bar', mark the areas that contain the object you want to detect using mouse and save it just by pressing 'S'. Utility will create the text file required fro training.
Utility can also be used to create sample images to be used for HMM training as well. If you are interested in HMM following http://www.bernardotti.it/portal/showthread.php?t=17777 has a good article on how to use HMM with OpenCV.