opencv_assignment3

Gesture recognition for camera and image

/*

•                            SPECIFICATION
  

• The programe aims to recognise gesture in an image. You can specify a static image
• for recognising, or capture image via camera without specifing any parameter.
• The processing workflow is
• #1 If there is a saved MLP 'bp.xml' under the same dir, load it
• #2 If there is no 'bp.xml' under the same dir, train the MLP
• #3 Preprocess the static image or frame from camera by following steps:

•  1) convert the image to gray
  

•  2) threshold, filter all pixels whose grey scale is larger than 230 or small than 70 to
  

•     avoid the affection by background
  

•  3) Median filter to avoid affection by noise
  

•  4) Threshold with specifing CV_THRESH_OTSU to extract the gesture
  

•  5) Find the largest contour
  

•  6) Calculate the CEs from the contour
  

•  7) Predict the result via the MLP
  

• #4 Show result

•                               NOTE
  

• #1 To train the MLP, there should be a folder named "trainImgs" under the same directory

• with the executable. All train images should putted be in the folder "trainImgs".
  

• #2 All names of all train images should contain at least 7 characters, and the 7th

• char should be the gesture number, for example, the train image
  

• "hand1_0_bot_seg_1_cropped.png" means the gesture of this image representes '0'.
  

• #3 The train process may take a few seconds to complete.
• #4 To recognise gesture via camera, you should make sure the background of the hand

• is black(or the grey scale is very different with the hand), so that the gesture
  

• could be extracted correctly.
  

*/

Any question, contact raoyonghui0630@gmail.com