基于二维投影变换的视频全景图拼接算法  

Video Panorama Mosaic Algorithm Based on 2D Projection Transform

作　　者：邹晓春[1] 何明一[1] 赵歆波[2] 

机构地区：[1]西北工业大学电子信息学院,西安710072 [2]西北工业大学计算机学院,西安710072

出　　处：《数据采集与处理》2012年第5期528-534,共7页Journal of Data Acquisition and Processing

基　　金：国家自然科学基金(60872144.61201319)资助项目;西北工业大学基础科研(JC20120205)资助项目

摘　　要：视频序列全景图拼接技术是视频图像分析的关键技术之一。针对帧间运动模型为二维投影变换的常见视频序列,提出了一种全景图拼接算法.该算法首先采用改进的基于RANSAC的特征点匹配算法对相邻帧的运动参数进行估计,接着采用直接法对各帧与全景图间的运动参数进行精确调整,最后运用自适应中值滤波绘制出全景图。实验结果表明,该方法能够生成较高质量的全景图,基本能实现无缝连接。Panorama mosaic is one of key techniques in video analysis. Aiming at the two-dimensional projective motion between frames used in the video sequence frequently, a mosaic algorithm for video sequence panoramic image is proposed. Firstly, the motion parameters of adjacent frame are estimated by using improved RANSAC-based matching algorithm for feature points. Secondly, the motion parameters between every frame with panorama are adjusted accurately by a direct estimation method. Finally, the video sequence panoramic image is plotted by self-adaptive median filter. Experimental results show that the proposed algorithm can generate the panoramic image with high quality.

关 键 词：视频处理 全景图像 RANSAC算法 运动估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]