Video stitching for surveillance applications

Public security has become a serious concern of governments. Therefore, many surveillance cameras are deployed in public areas by governments to monitor suspicious persons and objects. As the public areas are usually very large and the cameras have their limited views, security guards have to spot the suspicious persons and objects from many video screens. It is a very heavy work and dull process for the security guards and they need a new technology to let them watch the whole area in one screen. Video stitching technology could help solve this problem. The video stitching method could seam all video streams from cameras together into one video stream. In this dissertation, a video stitching method based on foreground extraction is proposed. Because moving objects would cause wrong registrations by directly using image stitching method on each frame from cameras and the poses of cameras are fixed, the homogeneous matrix or the transformation matrix is calculated from the background frames. This could reduce the computation complexity. The moving objects are separated from the frames and composited into the stitched background to eliminate the distortions caused by them. The moving objects are nearer to the cameras compared with the background, and the distance from it to the cameras could not be the same value. Hence, the scale ratios of the moving objects are normally larger than the background and could be different. In this dissertation, the scale ratios of the moving objects are adjusted to be the same as the background scale ratio. At last, a high resolution stitched video with an infinity viewpoint is obtained.Master of Science (Signal Processing