摘要
图像和视频拼接中,未引入全局变换的2幅图像或2段视频是无法拼接的。鉴于实际场景中的几何平面具有重要的结构信息,提出了一种基于平面区域匹配约束的图像视频拼接方法。利用深度卷积神经网络从RGB图像中提取语义信息,进而提取平面图像区域,将相邻视频画面间的平面区域含有的对应关系转化为约束,生成可用于指导图像拼接的优化网格,使得拼接结果在保持连续性的同时可平滑全局变换。该方法在现有语义分割网络中嵌入一个新模块,从而辅助平面区域语义分割。试验结果表明,该方法适用于大视差等场景。
In image and video stitching,it is difficult to stitch two images or two videos without intro⁃ducing global transformation.In view of the important structural information of geometric planes in re⁃al scenes,an image and video stitching method based on plane region matching constraint is proposed.The semantic information is extracted from RGB images by deep convolutional neural network,and then the plane image region is extracted.The corresponding relations between plane regions of adja⁃cent video images are converted in constraints,and an optimized network is generated to guide the im⁃age stitching,so that the stitching results can maintain continuity and smooth global transformation.In this method,a new module is embedded in the existing segmentation network to assist the semantic segmentation of plane regions.Experimental results show that this method is suitable for the large par⁃allax scenarios.
作者
赵志伟
张锐
贺敬武
ZHAO Zhiwei;ZHANG Rui;HE Jingwu(Department of Computer Science and Technology,Nanjing University,Nanjing 210023,China)
出处
《指挥信息系统与技术》
2022年第3期91-96,102,共7页
Command Information System and Technology
基金
“十三五”全军共用信息系统装备预研课题(31511040202)资助项目。
关键词
深度卷积神经网络
语义分割
平面区域
图像拼接
deep convolutional neural network
semantic segmentation
plane region
image stitching
