This is a basic gesture recognition software for the Kinect.
The hand is put in front of the camera, and the Kinect identifies which fingers are open and which are closed.

Each finger has a range of angles it can be from the center of the palm.
If, for instance, any fingertip found is in the angle region of the thumb, the thumb is considered open.



Dependencies:
-python (2.7+) (not python 3+)
-libfreenect
-opencv (3.0+) (for python)


To run, type into console:
  python main.py



Implemented features:
  -produce mask only of close objects
  -finds contour with area closest to area of hand, if within error bounds then considered to be hand
  -track palm of hand as circle, use as reference point on screen
    -find convexity defects of hand and bottommost point of hand
    -draw smallest encompassing circle around these points, is palm (see circles.py)
  -hand class
    -calibration function for hand
      -get contour polygon, points further from center than adjacent points on polygon = fingertips
      -highest point is middle finger, to left is index, etc.
      -get region that thumb is in when open, region that pinky is in when open, etc.
        -get angles of fingers from palm center, angles between fingers = boundaries to finger regions
        -store list of ranges of angles that each finger is within when open
          -eg. {thumb: [lowest ang of thumb, max ang of thumb], ...}
    -find all viable points on polygon to be finger
      -if far away enough from palm and angle to palm center is within a certain finger's angle region, that finger is considered open
      -if no finger point in that finger's angle region, that finger is considered closed
    -return dict of fingers and bools of whether they are open or not
  -HandStats class
    -augmented hand class, able to take samples of fingers and hand position at regular intervals over time
      -sampling allows for ignorance of random image noise
    -can use sampling to infer hand velocity and acceleration vectors
      -acceleration method is inaccurate (assumes acc is constant)
  -able to detect when no hand is present
    -Hand and HandStats methods to get open fingers, hand pos, etc. return None if no hand detected
    -no longer detects contours if beyond a certain distance away



Planned features:
  -deal with hand rotation to sides (?)
    -recalibrate if hand rotates a certain amount
  -gestures used to operate different programs
    -mouse
    -basic game (?)