Abstract
Growing interest arises in segmentation for object-based video clips since the development of MPEG-4 standard. The moving object extraction can also be applied to the object-based videoconference, surveillance, and so on. The difficulties of moving object segmentation are that physical objects are normally not homogeneous with respect to low-level features and it’s usually tough to segment them efficiently.The previous related researches are only operated with a static camera and in a stationary background. In this thesis, we propose a robust and fast segmentation algorithm and a reliable tracking strategy without knowing the shape of the object in advance. The system can segment the foreground from the background and track the moving object with an active (pan-tilt zoom) camera such that the moving object always stays around the center of images. Especially, the system can work in an unrestricted environment without the need for special purpose hardware.The proposed segmentation algorithm is based on the background subtraction, morphological operations, region growing, adaptive mechanism, template matching, and some innovative operations. The system can segment a moving object at 15 frames per second over a 176 x 144 pixel image.