Motion Detection In Multi-Camera Collaboration System

A multi-camera surveillance system for people detection and people tracking system is built in this thesis. The goal is matching and tracking the same target in different views of cameras.The mainly work is divided into two parts:1. People detection and tracking:We analyze HOG(Histograms of Oriented. Gradients)and SVM(Support Vector Machine)classifier for people detection first. Secondly, we give the detailed introduction of Meanshift algorithm and Kalman filter for people tracking.Followed by the performance evaluation on a standard video library.2. Multi-camera collaboration:There is a fixed geometric relationship among several certain cameras, which makes matching of the same object possible. The purpose of our work is to match, locate and identify different people in different camera views:Firstly, we use the epipolar geometry for people matching in two certain views; Then, trifocal tensor for people matching in three views is computed. Finally, Locating and tracking people in the third view is finished. A practical multi-camera experimental system is implemented finally.