Skip to content

Latest commit

 

History

History
131 lines (78 loc) · 3.53 KB

Instructions.md

File metadata and controls

131 lines (78 loc) · 3.53 KB

WIB project

author:tassosblackg | email:[email protected]*

OpenCV is required! I used OpenCV 3.1.0

Install OpenCV

To install see: https://docs.opencv.org/2.4/doc/tutorials/introduction/linux_install/linux_install.html

and this:https://www.pyimagesearch.com/2016/10/24/ubuntu-16-04-how-to-install-opencv/

|possible error if you use the above link to solve check this: https://github.com/opencv/opencv/issues/6050

Run

  • Steps to run classification and detection of ball

    1. You need a folder with imgs for negative training (negativeImg)

    2. You need as many positive imgs by croping the obj (showing the obj you want to recognize) that's /BallImages

  • Starting to build necessary components

    cd  /files

    > read --README.md

    1. follow steps of /files/README.md file to create bg.txt and ball.info file

      bg.txt will be used for creating samples and training..so is important!

    2. cd back to the main project folder (/W.I.B) | cd ../

IMPORTANT :

Run steps (6->8) only if you want to retrain the classifier with your datashet/

First make sure to delete the existing directory /data

|rm -rf /data

Delete also ball.info,ball.vec,bg.txt

| rm   *ball.info,ball.vec,bg.txt*

  • Creating samples utility [Creating training set]_

    opencv_createsamples  -info ball.ifo  -num 40 -w 16 -h 16 -vec ball.vec 

      arguments:

     -**num**, *is the number of positive samples you have in the ball.info file.*

-**w** & -**h**, *the wanted size of useful object.*

-**vec**, *is the binary file that will be used for opencv_traincascade utility.*

Notice: Alternative you can run, opencv_createsamples -vec ball.vec -w 16 -h 16 , to see the created sample use keyboard arrows to change between images

  • opencv_traincascade

    mkdir data

    opencv_traincascade -data data -vec ball.vec -bg bg.txt -numPos 30 -numNeg 80 -numStages 5 -w 16 -h 16 -featureType LBP

    arguments:

    -**data**, *is the folder where training-cascade .xml files are stored*

-**bg**, *is the background file with images' path*

-**numPos**, *is the number of positive images to use for training per stage (GENERAL_RULE : numPos < num) 

   |where num, see creating_samples

-**numNeg**, *is the nuber of negative images are stored in bg.txt*

-**numStage**, *is the number of stages that we want our cascade to have*

-**w** & -**h**, *are the same as before*

-**featureType**, *is the type of cascade we want LBP or Haar*

  • Build detect ball (see /src , wib.cpp, src/README.md)

    Last but not least..

    9) 

```
*cd  /src
```
  >  read README.md and build the .cpp code*

    ./wib img_file_to_detect

For more information about OpenCV cascade classifier check :https://docs.opencv.org/2.4.13/doc/user_guide/ug_traincascade.html