摘要
本文针对传统的低空无人机影像拼接处理速度慢和几何精度低的问题,提出了一种快速的无人机影像无缝拼接方法:对输入的原始影像按一定尺度进行降采样,在降采样的影像上进行SURF特征提取和匹配,利用RANSAC方法估计初始的相对单应矩阵,然后用Levenberg-Marquardt方法精化单应矩阵,计算出初始的绝对单应矩阵后利用稀疏光束法平差估计出精确的单应矩阵,通过降采样影像与原始影像的单应关系传递单应矩阵到原始影像级,最后进行影像合成,形成整体拼接图。实验结果表明该方法可以有效地提高拼接速度,解决拼接错位问题。
Since the traditional mosaicking of low attitude unmanned aerial vehicle （UAV） images takes highly computational cost and low geometric precision, a fast method for image registration and seamless mosaicking was proposed in the paper. First, original images were downsampled, then the SURF algorithm was used to extract point features, and image matching was implemented. Sec- ond, initial homography matrix between adjacent images was computed by RANSAC method, then refined by Levenberg-Marquardt method, and the final absolute homography matrix was determined by sparse bundle adjustment. At last, the absolute homography ma- trix was transformed to the original image by down-sampling homographic relations and image mosaic was done. The experimental resuits showed that this mosaic method performed fast and could solve the problem of misregistration well.
出处
《测绘科学》
CSCD
北大核心
2012年第5期23-26,共4页
Science of Surveying and Mapping
关键词
影像拼接
无人机
稀疏光束法平差
SURF
单应矩阵
拼接速度
image mosaic
UAV
sparse bundle adjustment
Speeded Up Robust Feature （SURF）
homography matrix
mosaic speed
