why implementing mixture models for detection turned out to be a nightmare

Recently I have spent a large chunk of my time implementing the adaptive mixture model paper written by Stauffer and Grimson from 1998.  It is an old paper which introduces the idea of modeling the value of a pixel over time with K Gaussian components. You would expect that most of the time a pixel has an intensity corresponding to the background scene, and some of the time the pixel’s intensity jumps indicating a foreground object. If we can learn the background distribution quickly then we can treat pixels that fall inside the background distribution as such and those that don’t fall inside are classified as foreground. When I read this paper the first time I thought it was great, it seemed adaptive enough to deal with ‘real-life’ challenging data. I was optimistic and decided to implement this algorithm in Matlab which shouldn’t take that long right? It was after working on the implementation that I realized I actually hated the paper, and hated implementing the algorithm.

SVN Version Control Repository for an Existing Project

There are times when I want to keep track of the changes I make to the code I’m writing. For instance right now I have implemented the Stauffer and Grimson paper on using mixture of gaussians for foreground / background detection. As I make changes to the algorithm itself I want to keep track of the changes, that way if I want to go back and see what my algorithm did before I can do that.

So in this post I will quickly describe how you can begin tracking changes (version control) for an existing project using SVN Tortoise for Version Control.