Using Iterative Closest Points (ICP) for tracking partially occluded shapes

This is an application of the ICP from github.com/Cla... in the context of my thesis (work in progress, early stage).
Explanation: The input image on the top is processed with OpenCV in order to detect the hand and the different shapes. A depth cam is used to find out when the hand touches the table. When a shape is touched, a snapshot of its contour is made (white contour in the bottom left window). Then the picture is filtered in order to isolate the shape that is being touched (mostly by HSV color filtering), visualized in the bottom right window. I then sample up to 500 random points on both the snapshot contour and the recently detected contour that is partially occluded by the finger. Finally, ICP is run on the partially occluded shape with respect to the snapshot contours. The transform matrix is applied to each point sampled on the snapshot contours, resulting in a prediction of t
Processing is real-time on a i7-4790K.

[@emmonsmckinney3249]

neat!