In this thesis we tried to exploit Dominant set frame work to track multiple people in videos. Since it's unsupervised, given unlabeled patches of the people to follow, we then propose a formulation of the multi-target ...