Real time hand keypoint detection through motion

A simple program I made with traditional computer vision techiniques in OpenCV. Unlike every other hand detection method I found on github, you do not need to isolte your hand against a plain wall. It works by using background subtraction to find keypoints instead of simple image transforms to a binary image. Check out demo video at bottom.

1) Image transform

Mirror, convert to grayscale, and blur webcam frame

2) Perform Background subtraction.

You are left with a binary image of the estimated foreground but the image is sparse

3) Morphological transformations in order to fill in sparsity and remove noise

4) Find contours and then convex hulls of morpholocial output

Provides final output of keypoints (good for fingertip detection), only detectable when moving hand

5) Game mechanics

Draw a circle and have it switch direction when hitting a wall or coming in contact with a hand keypoint. In order to detect collision, use pointPolygonTest function to check if a point within the radius of the circle (10 degree incriments) is within any of the convex hull keypoints.

Demo Video