بهنام رحیمی کلاریجانی

In this paper an automated and robust method for detection and tracking of moving target with assumption of camera mobility using scale invariant feature transform is proposed. By automated registration technique of current frame on the previous frame, a panorama based background model of scene is generated. Then for real-time panorama-based background subtraction, frame registration into the panorama and global homographic matrix estimation are used. After that for each new frame, moving targets are detected and background model is updated and the key points where detected in the boundary region around the target are used for tracking. By matching description vectors of at least three key points where extracted from scale invariant feature transform of current frame to previous frame, initial estimation of target location on current frame are obtained. In the next step, more accurate estimation of target location on current frame is obtained by using Kalman filter. This approach is able to be used in different statues, such as variable climate condition, closing or faring the target, partial and global occultation. Comparison to expert operator, this approach improved performance results and achieved high accuracy in detection and tracking of aerial and ground target.