This repository aims to provide tools and information on training your own OpenCV Haar classifier. Use it in conjunction with this blog post: Train your own OpenCV Haar classifier.
-
Install OpenCV & get OpenCV source
brew tap homebrew/science brew install --with-tbb opencv wget http://downloads.sourceforge.net/project/opencvlibrary/opencv-unix/2.4.9/opencv-2.4.9.zip unzip opencv-2.4.9.zip -
Clone this repository
git clone https://github.com/mrnugget/opencv-haar-classifier-training -
Put your positive images in the
./positive_imagesfolder and create a list of them:find ./positive_images -iname "*.jpg" > positives.txt -
Put the negative images in the
./negative_imagesfolder and create a list of them:find ./negative_images -iname "*.jpg" > negatives.txt -
Create positive samples with the
bin/createsamples.plscript and save them to the./samplesfolder:perl bin/createsamples.pl positives.txt negatives.txt samples 1500\ "opencv_createsamples -bgcolor 0 -bgthresh 0 -maxxangle 1.1\ -maxyangle 1.1 maxzangle 0.5 -maxidev 40 -w 80 -h 40" -
Compile the
mergevec.cppfile in the./srcdirectory:cp src/mergevec.cpp ~/opencv-2.4.9/apps/haartraining cd ~/opencv-2.4.9/apps/haartraining g++ `pkg-config --libs --cflags opencv | sed 's/libtbb\.dylib/tbb/'`\ -I. -o mergevec mergevec.cpp\ cvboost.cpp cvcommon.cpp cvsamples.cpp cvhaarclassifier.cpp\ cvhaartraining.cpp\ -lopencv_core -lopencv_calib3d -lopencv_imgproc -lopencv_highgui -lopencv_objdetect -
Use the compiled executable
mergevecto merge the samples in./samplesinto one file:find ./samples -name '*.vec' > samples.txt ./mergevec samples.txt samples.vec -
Start training the classifier with
opencv_traincascade, which comes with OpenCV, and save the results to./classifier:opencv_traincascade -data classifier -vec samples.vec -bg negatives.txt\ -numStages 20 -minHitRate 0.999 -maxFalseAlarmRate 0.5 -numPos 1000\ -numNeg 600 -w 80 -h 40 -mode ALL -precalcValBufSize 1024\ -precalcIdxBufSize 1024 -
Wait until the process is finished (which takes a long time — a couple of days probably, depending on the computer you have and how big your images are).
-
Use your finished classifier!
cd ~/opencv-2.4.9/samples/c chmod +x build_all.sh ./build_all.sh ./facedetect --cascade="~/finished_classifier.xml"
A huge thanks goes to Naotoshi Seo, who wrote the mergevec.cpp and
createsamples.cpp tools and released them under the MIT licencse. His notes
on OpenCV Haar training were a huge help. Thank you, Naotoshi!