本文对视频拼接设计了一个简单的框架，主要是采用三路固定摄像进行实时拼接，在Uuntu16.04 系统安装 OpenCV开源库和CUDA8.0 软件包，并使用该系统搭建视频拼接框架系统。利用GPU编程技术，对视频拼接技术中最耗时 的特征匹配算法、细缝估计、多波段融合和扭曲映射算法进行加速。在计算资源足够的情况下，利用CUDA进行在线实时视 频流拼接。

This paper designs a simple framework for video stitching, mainly using three fixed camera for real-time stitching; installs OpenCV base and CUDA8.0 software package in the Uuntu16.04 system; and uses the system to build video stitching frame system. It uses GPU programming technology to accelerate the most time-consuming algorithms of video stitching technology, such as feature matching, slit estimation, multi-band fusion and twist mapping. With sufficient calculation resources, CUDA is used for the online real-time video stream stitching.