OutletDetector

This is an example of training a boosted cascade of weak classifiers for electrical outlet detection using OpenCV. The official document is here.

Requirements

• OpenCV=3.2.0 (OpenCV 4.x has removed cascade trainer applications. I also tried OpenCV 3.4.0, however, the result is not as good as that of 3.2.0. )

• imgaug

Usage

1. Install requirements

   Download opencv 3.2.0 release package. Extract and add %your_opencv_root%\build\x64\vc14\bin to PATH. If you are using linux, you need build the source by yourself. Remember to select BUILD_APP option.

   Install python requirements

   pip install -r requirements.txt
   

2. Augment positive example

   python augmentation.py
   

   The positive examples will be stored in dir 'p'.

3. Create positive and negative list

   python create_lst.py
   

   This will creates pos.lst and neg.lst.

4. Start training

   If you want to detect the same outlet pattern as mine, you can directly use the trained xml presented. Otherwise, you need first empty the path 'classifier' and run train.bat (or change to train.sh on linux).

   train.bat

   opencv_createsamples -vec outlet.vec -info pos.lst -num 401 -w 24 -h 24
   opencv_traincascade -data classifier -vec outlet.vec -bg neg.lst -numPos 380 -numNeg 300 -numStages 15 -w 24 -h 24
   

   The meaning of parameters can be found in the official document. It should be noted that the -numPos should be set somewhat less than -num to make training successful.

5. Test your model

   python haar_find.py
   

   You can select detecting in image or in webcam stream by switching mode in haar_find.py.