以卫星影像为底图的无人机图像自动拼接方法  

An UAV image automatic stitching method taking satellite image as base map

作　　者：侯天怡 尤红建[1,2] 孙显[1,2] 

机构地区：[1]中国科学院空间信息处理与应用系统技术重点实验室,北京100190 [2]中国科学院电子学研究所,北京100190 [3]中国科学院大学,北京100049

出　　处：《测绘科学》2015年第6期11-16,35,共7页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41001285)

摘　　要：针对无人机影像受成像角度倾斜、光照不均匀等影响,导致常规匹配方法拼接效果较差的问题,该文提出一种基于现有卫星影像的由粗到精无人机图像自动拼接方法。首先,在对图像进行粗匹配阶段,提出了一种基于环状描述子的尺度不变特征变换,用于解决图像间旋转和尺度的差异,较好地克服了SIFT算法描述子维度较高且计算量大的不足。其次,引入分块互信息进行精匹配,减少拼接过程中的累计误差。最后,采用该文提出算法对无人机图像进行拼接,实验结果表明,该文提出的无人机图像拼接算法在满足实时性要求的同时可以达到较高精度。Aiming at the problem that general image stitching methods always perform poorly since un-manned aerial vehicle（UAV） images are affected by flight attitude and uneven illumination, this paper putforward a coarse-to-fine automatic image stitching algorithm based on existing satellite image. First, a ringdescriptor based scale invariant feature transform algorithm was proposed to improve the shortage of SIFTdescriptor＇s high dimensions and large computation. The improved algorithm was used for image coarsematching corrected image rotation and scale differences. Second, the accumulated error was reduced bytaking advantage of block mutual information algorithm. Experimental results showed that the proposedimage stitching algorithm could meet real-time requirements and achieve high accuracy at the same time.

关 键 词：图像拼接 环形尺度不变特征变换 互信息 异源图像匹配 卫星影像 无人机 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]