An Efficient Video Alignment Approach for Non-Overlapping Sequences with Free Camera Movement

This paper describes a method to align two videos. The idea can be extended for alignment of multiple video sequences. The sequences are recorded by un-calibrated video cameras with fixed (but unknown) internal parameters. It is shown that by using a combination of spatial information, temporal changes and the inbuilt frame-to-frame transformation of the video sequences, efficient video alignment can be achieved. The sequences used in the algorithm described may have no or minimal spatial overlap and the camera sources are also allowed free movement in space. Video alignment gives rise to a wide variety of new applications that are not possible when only image-to-image alignment is used