Title: The Horn-Schunck method for optical flows
Speaker: Jon Eivind Vatne, HiB
Time and place: Wednesdays 16.09 and 30.09, 14.15–15.00 in E203
Abstract:
Optical flow is a way to capture movement in a video sequence, or the difference between the pictures. Horn and Schunck published a method for computing optical flow around 1980. The basic steps for setting up the algorithm are: Starting from an underdetermined system of equations, introduce an L^2 penalty to turn the problem into a minimization procedure. Using the calculus of variations, the problem is turned into a system of partial differential equations. The problem is then discretized and solved using standard numerical methods from linear algebra.
We will discuss the set-up and the proof of convergence of the numerical solution scheme.
Part 2: In part 1 we ended with the system of partial differential equations. In part 2, after a brief recap, we will consider the discretization and convergence of the iterative solution.