Human tracking using surveillance cameras is a demanding topic of research now-a-days. Tracking and recognizing the human is much more challenging. There are many existing methods for tracking humans based on shapes and motion. In this paper, a novel algorithm for tracking human under occlusions is introduced. Gaussian Mixture Model (GMM) is used for tracking the human, which performs well under different occlusions. The new algorithm produces excellent results in case the human is occluded by another human and is obstructed by some other human, and also when there are partial occlusions. Experimental results show that the new algorithm outperforms in tracking the humans in these three cases.