Today millions of video surveillance are installed throughout the world. The stated goal of installation of CCTV cameras is to increase public safety. Automatic video analysis system for the large number of video is a current problem. This thesis aims at the problems of person detection, tracking and especially re-identification in a non-overlapping camera network. It’s a difficult task due to some challenges like viewpoint variation, person pose variation, low resolution and so on. In this thesis, we plan to use the person semantic attribute to aid the re-identification and apply the deep learning methods lile a Siamese network in order to learn the similarity metric between two images.