An adaptive seam line panoramic video stitching algorithm

Abstract

In order to improve the phenomenon of artifacts and loss when moving objects pass through the overlapping area in panoramic video stitching, an adaptive panoramic video stitching algorithm is proposed by combining the characteristics of panoramic video stitching. Firstly, extracting video frames, computing overlap areas, and the calculation method of the energy function and the search direction of the seam line are improved based on the traditional dynamic planning algorithm to find the seam line. The traditional method only judges the three adjacent points in the previous row of the current pixel point instead of all the pixel points in the previous row, and this improved method is used to find multiple seam lines in the overlapping area, detect the moving object target, and update the seam line in time. The experiments show that the improved algorithm can not only avoid the larger objects in the overlapping area when finding seam lines, but also improve the artifacts and loss caused by moving objects passing through the seam lines in the overlapping area in video stitching. At the same time, the GPU processing achieves real-time effect.