Abstract: In the video stitching, registration error and moving objects lead to ghost error. It brings about the loss of high-frequency content by simple fusion method, and complex fusion algorithm is hard to meet real-time fusion. To solve these problems, this paper uses the improved Harris feature operator, and the improved RANSAC algorithm to reduce image registration error. It combines with the bundle adjustment and puts forward a new fusion method. Through the overlapping regions of the border to strike and morphology corrosion science operations, it determines the need for integration of the pixel, and then as a weighted combination of trigonometric functions to carry out integration. A large number of experimental results show that this method has a good effect of real-time stitching.

Key words: video stitching, ghost error, fusion, morphology, bundle adjustment

摘要： 在视频拼接中，配准的误差和运动物体都会给拼接结果带来鬼影，简单的融合方法会带来图像高频内容的丢失，复杂的融合算法则难以满足实时性。针对上述问题，使用一种改进的快速检测Harris算子以及改进的RANSAC算法来减少图像配准误差，结合捆绑调整进行全局调整，并提出新的融合方法。通过对重叠区域边界的求取和形态学腐蚀运算，确定需要融合的像素，结合三角函数作为权值来进行融合。大量的实验结果表明，该方法具有良好的实时拼接效果。

关键词: 视频拼接, 鬼影融合, 形态学, 捆绑调整

CLC Number: 

• TP391