无控制点场景下基于多运动目标的卫星视频稳像算法  

Satellite video stabilization algorithm based on multiple moving targets in no⁃control⁃point scenarios

作　　者：李睿瑄 李峰 辛蕾 杨雪 张南 LI Ruixuan;LI Feng;XIN Lei;YANG Xue;ZHANG Nan(Qian Xuesen Laboratory of Space Technology,Beijing 100094,China)

机构地区：[1]钱学森空间技术实验室,北京100094

出　　处：《计测技术》2024年第2期70-81,共12页Metrology & Measurement Technology

基　　金：国家自然科学基金项目(62205371);国家重点研发计划项目(2020YFA0714101)。

摘　　要：为了解决无控制点场景下现有卫星视频稳像算法失效的问题,提出了一种基于多运动目标的卫星视频稳像算法,该算法构建了目标检测、轨迹平滑和范数优化的框架,将多目标卡尔曼滤波、rlowess轨迹平滑与L1-L2范数优化相结合,最终实现卫星视频稳像。利用海上多目标舰船观测数据集开展实验,结果表明基于多运动目标的卫星视频稳像算法在X、Y方向上的稳像误差均不超过0.3个像素,验证了该算法的有效性。本研究填补了无控制点场景下卫星视频稳像的技术空白,具有重要的工程应用价值。To address the failure of existing satellite video stabilization algorithms in no⁃control⁃point scenarios,a satellite video stabilization algorithm based on multiple moving targets is proposed.This algorithm constructs a framework that integrates target detection,trajectory smoothing,and norm optimization by combining multi⁃target Kalman filtering,rlowess trajectory smoothing,and L1⁃L2 norm optimization,ultimately achieving satellite video stabilization.Experiments were conducted using a maritime multi⁃target ship observation dataset.The results show that the stabilization error of the proposed algorithm can be controlled within 0.3 pixels in both the X and Y directions,and validate the effectiveness of the algorithm.This study fills the technical gap in satellite video stabilization in no⁃control⁃point scenarios and has broad and significant engineering application value.

关 键 词：无控制点场景 卫星视频稳像 多运动目标 亚像元级 卡尔曼滤波 rlowess平滑 L1-L2范数优化 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置] TB9[自动化与计算机技术—控制科学与工程] V19[一般工业技术—计量学]