Abstarct: People Behavior analysis has lots of applications in various fields including customer behavior analysis in shopping centers and protection of facilities such as banks and airports. Tracking people is the first step in such systems. It is not possible for a single camera to observe a wide environment due to the limitation of sensors. Therefore a network of cameras is needed. In addition of Single camera tracking problems, Multi camera systems especially non-overlapping views encounter with other challenges. For example appearance of people is changeable due to situations such as different illumination, position of people, view angle and other parameters of the camera. In addition presence of people in camera views doesn’t follow a certain rule and they can widely separate in time and space especially when cameras have no common area in their views, which make the problem more challenging. In this thesis, a new method is proposed to track people across disjoint camera views for an indoor environment. In this method, tracking people is performed in two stages. In the first stage people are tracked in each camera independently. Occurred events in each camera are saved in a central databaxse with a standard data structure. In the second step, each person is tracked in whole environment, baxsed on the information saved in databaxse. Background modeling and frxame difference is used in order to detect people in a single camera. After applying pre-processes to improve the quality of detected objects and eliminating disturbing thing such as shadows, trajectory of each person is determined by Kalman Filter and MHT. Proposed method can track people in different illumination conditions and in presence of occlusions. In order to establish correspondence between observations, which are related to a person, across cameras, Extracted features are compared with those that are saved in databaxse. In this thesis a new feature is proposed that is more robust in different illumination conditions as well as being more distinguishable. This feature utilizes the main diagonal elements of Co-Occurrence matrix in YCbCr color space. Proposed method is tested on samples from 5 cameras with different views of a building. Results showed that this feature is more robust than features like color.