Train your own OpenCV Haar classifier

Important: This guide assumes you work with OpenCV 2.4.x. Since I no longer work with OpenCV, and don't have the time to keep up with changes and fixes, this guide is unmaintained. Pull requests will be merged of course, and if someone else wants commit access, feel free to ask!

Instructions

1. Install OpenCV

2. Clone this repository

3. Put your positive images in the ./positive_images folder and create a list of them:

    find ./positive_images -iname "*.jpg" > positives.txt
   

4. Put the negative images in the ./negative_images folder and create a list of them:

    find ./negative_images -iname "*.jpg" > negatives.txt
   

5. Create positive samples with the bin/createsamples.pl script and save them to the ./samples folder:

    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"
   

6. Use tools/mergevec.py to merge the samples in ./samples into one file:

    python ./tools/mergevec.py -v samples/ -o samples.vec
   

   Note: If you get the error struct.error: unpack requires a string argument of length 12 then go into your samples directory and delete all files of length 0.

7. 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
   

   If you want to train it faster, configure feature type option with LBP:

     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 -featureType LBP
   

   After starting the training program it will print back its parameters and then start training. Each stage will print out some analysis as it is trained:

   ===== TRAINING 0-stage =====
   <BEGIN
   POS count : consumed   1000 : 1000
   NEG count : acceptanceRatio    600 : 1
   Precalculation time: 11
   +----+---------+---------+
   |  N |    HR   |    FA   |
   +----+---------+---------+
   |   1|        1|        1|
   +----+---------+---------+
   |   2|        1|        1|
   +----+---------+---------+
   |   3|        1|        1|
   +----+---------+---------+
   |   4|        1|        1|
   +----+---------+---------+
   |   5|        1|        1|
   +----+---------+---------+
   |   6|        1|        1|
   +----+---------+---------+
   |   7|        1| 0.711667|
   +----+---------+---------+
   |   8|        1|     0.54|
   +----+---------+---------+
   |   9|        1|    0.305|
   +----+---------+---------+
   END>
   Training until now has taken 0 days 3 hours 19 minutes 16 seconds.
   

   Each row represents a feature that is being trained and contains some output about its HitRatio and FalseAlarm ratio. If a training stage only selects a few features (e.g. N = 2) then its possible something is wrong with your training data.

   At the end of each stage the classifier is saved to a file and the process can be stopped and restarted. This is useful if you are tweaking a machine/settings to optimize training speed.

8. 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).

9. Use your finished classifier!

    cd ~/opencv-2.4.9/samples/c
    chmod +x build_all.sh
    ./build_all.sh
    ./facedetect --cascade="~/finished_classifier.xml"
   

Acknowledgements

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!

References & Links:

• Naotoshi Seo - Tutorial: OpenCV haartraining (Rapid Object Detection With A Cascade of Boosted Classifiers Based on Haar-like Features)
• Material for Naotoshi Seo's tutorial
• OpenCV Documentation - Cascade Classifier Training