Abstract

To meet the growing requirements of different video applications such as video surveillance or coding, many video processing techniques have been developed to analyze and represent video sequences. Video object segmentation is an object-based video processing technique which aims to detect semantically meaningful components, i.e., objects, in a video sequence. In case the video sequence contains global (camera) motion, global motion estimation is required to compensate the camera motion before segmentation. This thesis studies methods to automatically segment moving objects in the presence of camera motion without user interaction. It proposes a fast and robust global motion estimation method oriented to video object segmentation. In addition, it integrates this method into a modular scheme to segment objects in the presence of camera motion. This video object segmentation scheme consists of three main steps: global motion detection, global motion estimation and compensation, and object segmentation. The object segmentation is based on: change (motion) detection, temporal adaptation, and edge adaptation. Some improvements are proposed in each part of the object segmentation. The proposed methods aim at four goals: automatically adapt to camera motion, robust (insensitive) to noise and artifacts, temporally stable segmented objects and low computational cost. The proposed methods are reliable which was confirmed by experimenting on more than 10 indoor and outdoor video shots both with and without camera motion. Simulation results show that the proposed GME method achieves more satisfactory results than the reference methods. For object segmentation, encouraging results are also achieved.