Line detection and timestamps, video, Python

edit

Lines and faces are completely different objects, so you shouldn't count on facedetect.py. Try to start from the Hough transform. You can grep OpenCV C/C++ samples for this function, there should be some useful code samples.

HoughLines should be a perfect function for pencils on paper, but melting ice will produce curves, not strict edges. Nevertheless you can start with this function, and switch later to tracking with Optical Flow or something. You can put several dozens of points to every edge, and then track how they move down. But you should constrain the Optical Flow in a way so it searches for new points' positions in the vertical direction only. If the ice is melting slowly, you can even implement your own search with a block-matching-like technique.

If your paper background is quite plain, you can do some simple binary thresholding to get the pencil lines. Then projecting the lines onto the y axis and doing a min for a range will give you the lowest point of a particular pencil line and you can set some predefined points in y that you are looking for to be logged. Whenever the min(projection(y)) is lower than your predefined line, you trigger the rest of the code. However the assumption is that the various pencil lines do not overlap in the projection space.