With the popularization of vision-based robots and video camera applications, object tracking became one of the major fields in computer vision system. While the other two major fields, recognition and detection, basically operate on one image frame, tracking algorithms deal with a set of consecutive frames that information from previous frames are also taken into account. Changes in object directions, scales and rotations between frames, computation time and occlusions are major difficulties in object tracking and one of the popular methods robust to these problems is Mean-shift tracking. In this thesis, tracking procedure was divided into two different modes of Target tracking and Target searching. Tracking was processed with color-based Mean-shift tracker with Target-adjusted color model and new window adjusting schemes. Searching process was processed with SIFT features matching. A new Mean-shift tracking strategy robust to unexpected movements of target objects or viewers was proposed. While reducing the probability of falling into erroneous states, it still maintains its advantages-a small number of computations by focusing on local information around the target and tracking of non-rigid objects. Target searching process is executed when the similarity between the current region and the previously saved target region is low, that is when the initial state for the Mean-shift is not yet proper. The new initial location of the tracking window is computed from the SIFT features matching result. Note that this searching is also done in a local region, the size of the region being chosen according to the value of a similarity coefficient. The purpose of this thesis is to propose and implement a new tracking method especially robust to abrupt motions or scale changes. To satisfy this, using color-based tracking and feature-based searching was proposed, a new window adjusting schemes was proposed and the Target-adjusted color model was used. The usefulness of thi...