You've been asked to implement Lucas-Kanade tracking, and as your extra you should implement Gaussian windowing as in the wikipedia article. Several questions:

- In our previous assignment we took great care to make sure the size of the filter in the Gaussian was wide enough to capture all sample points where the Gaussian was above a threshold. Then we normalized the values so they summed to 1. Do we need to do these steps for Gaussian windowing for tracking? Why or why not?
- In Assignment 6 you are asked to track a point for several frames, experimenting with various settings of Gaussian smoothing and window size. Perform these also with Gaussian windowing but here vary the sigma for the window, and report your results.
- Let's consider what happens when you try to track a highly textured surface, when its projected motion in the image is fast, and when it is slow. Should we change the amount of smoothing we apply for the two cases, before we take derivatives? Why or why not? Hint: recall that the optical flow constraint equation is just a first-order expansion of the derivative at a point.