节点文献
基于H.264的全景视频编解码技术研究与实现
The Research and Implementation of Panoramic Video Encoding and Decoding Technology Based on H.264
【作者】 李伟;
【导师】 曾文献;
【作者基本信息】 河北经贸大学 , 计算机应用技术, 2018, 硕士
【摘要】 随着智慧旅游、虚拟现实、无人驾驶等多媒体技术及应用的不断发展,传统视频已无法满足要求,全景视频以其全视角特点成为近年研究的热点。然而受摄像头采集视角有限等因素的限制,无法自动获取全景视频,目前,全景视频主要由多路视频拼接而成,因此多路视频帧同步、拼接矩阵计算、快速编码等技术是生成全景视频的关键点。全景视频由一系列连续的全景图像组成,本文对全景视频生成过程中最耗时的全景图像生成和运动估计展开研究。全景图像生成主要选用图像拼接技术,使用32维SIFT算法构建初始拼接矩阵映射表,基于矩阵映射表实现了快速生成全景图像;为提高全景图像的准确性,提出了基于BP神经网络的拼接矩阵自适应调整方法,不断对矩阵映射表中拼接矩阵进行自适应调整,实现了全景图像快速、准确生成。分析了H.264视频编码中运动估计环节,提出了螺旋式运动估计方法,通过运动矢量预测确定匹配块搜索的起始点,然后利用双阈值配合可扩充螺旋式搜索模板和小菱形搜索模板实现了最优运动矢量的快速计算。通过实验,验证了基于神经网络的全景图像快速生成方法和螺旋式运动估计方法的可行性。开发了全景视频生成及播放系统,证明了方法的实用性。
【Abstract】 According to the progress of time and the development of smart tourism,virtual reality,self driving and other multimedia technology and applications,traditional video can’t satisfy people’s necessity.Panoramic video has become a hot research topic in recent years because of its full view characteristics.However the camera view is limited and other factors,can automatically obtain the panoramic video,the panoramic video mainly consists of multiple video splicing,so multi-channel video frame synchronization,splicing matrix calculation,real-time encoding technology is the key point for generating panoramic video.Panoramic video is composed of a series of continuous panoramic image frames.This paper studies the most time-consuming panoramic image generation and motion estimation in panoramic video generation.Panoramic image generation mainly uses image mosaic technology,using the initial registration matrix mapping table improved fast SIFT algorithm for generating matrix,mapping table can be generated quickly based on panoramic images;to improve the accuracy of the panoramic image stitching,proposed matrix adaptive adjustment method based on BP neural network,continuously adjust the connection matrix mapping table in the matrix,to make rapid and accurate generation of panoramic image.The H.264 motion estimation in video encoding process are analyzed,a method is proposed to estimate the spiral motion vector prediction by determine the matching block of the initial search point,using double threshold with expandable screw type and small diamond search template to achieve a rapid calculation of the optimal motion vector search template.Through experiments,the feasibility of the fast generation method of panoramic image and the spiral motion estimation method based on neural network is verified.A panoramic video generation and playback system is developed to prove the method.
【Key words】 Panoramic video; image stitching; H.264; BP neural network; SIFT;