Video image stitching based on moving object detection and motion prediction compensation

This paper proposed a novel image stitching algorithm based on fast moving object detection and motion prediction compensation, which can avoid ghosting caused by the moving target. We estimates the global motion vector through extracting and matching SIFT features. Based on the global motion estimation, calculate the relative difference image of the overlapping region in the pair of registered images. Then the gray-scale projection curve of the difference image is analyzed to judge whether moving object exists or not. If a moving object appears, we dedicate to updating the moving object with relevant background information collected from the next frame or the one currently reserved in buffer. The experiments show good performance for stitching video sequences in the presence of moving object.