how can gaussian mixture model segment foreground of an image?what is the working principle behind it?

edit

I am working on person tracking project on videos where I am going to extract foreground objects i.e persons using gaussian mixture model based function. After segmentation I m going to track them based on the object's movements. So i need theory behind gmm working. I found info about it in pure, deep statistical level but i need easily understandable explanation of it applied to image segmentation along with its statistical background.