基于特征的航空视频序列拼接方法  

Feature-based mosaic method for aerial video sequences

作　　者：王家宝[1] 张亚非[1] 陆建江[1] 周波[1] 徐伟光[1] 李阳[1] 

机构地区：[1]解放军理工大学指挥自动化学院,江苏南京210007

出　　处：《解放军理工大学学报（自然科学版）》2012年第2期159-164,共6页Journal of PLA University of Science and Technology(Natural Science Edition)

基　　金：国家部委预研基金资助项目(9140A06050409JB8102)

摘　　要：针对航空侦察视频序列,提出了一种基于特征点的视频序列拼接方法,论述了该方法的流程和模块。该方法矫正了航空视频序列中普遍存在的径向扭曲;改进了SIFT特征提取和匹配策略,缩短了算法的时间;提出了改进的RANSAC算法用于剔除匹配外点,提高了视频帧图像之间的变换精度;提出双频带融合技术消除边缘跳变和模糊叠影现象,改善了视频帧融合的视觉效果。实验结果表明,该方法能够有效地实现航空视频序列的近实时拼接,同时保持对噪声、光照和目标运动鲁棒。A feature-based mosaic method was proposed for aerial video sequences,and the flow and modules of the method were addressed.The pervasive radial distortion in aerial video sequences is rectified.The methods of extraction and matching of the SIFT features are improved,and the time of the algorithm is saved.The outliers are kicked out by this improved RANSAC,which enhances the precision of the transformation among frames.Finally,the visual effect of mosaic is increased by the proposed double-band blending technique,which eliminates the discontinuous edge and blurring.Experiments prove that the method could realize mosaic in nearly real time,and is robust to image noise,illumination variations and moving objects.

关 键 词：航空视频 图像拼接 径向扭曲 SIFT特征 RANSAC算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]