三维重建中的大规模航空影像检索方法

Large scale aerial image retrieval method in 3D reconstruction

针对互联网获取的航空影像数据或非摄影测量专业人员拍摄的无人机影像数据进行高精度测绘与三维重建时,从中检索相似影像较为困难的问题,该文提出了一种基于尺度不变特征变换(SIFT)算子与词汇树的大规模航空影像检索方法。该方法可以从规模庞大、排序复杂的航空影像中快速准确地检索出相似影像。首先利用SIFT描述子来代表影像,然后通过分层K-means算法对影像库的SIFT描述子构建词汇树,最后再利用TF-IDF方法加权计算待查询影像与影像集中各影像的相似度,获得相似影像。通过对多组大规模航空影像数据进行试验,结果证明了该文所提算法的可靠性、准确性和实用性。

Aiming at the problem of how to retrieve images with overlapping regions quickly and accurately while using these image data obtained from the Internet or non-photogrammetry professionals in unmanned aerial vehicle(UAV)imaging data for high-precision mapping and 3 Dreconstruction,this paper proposed a large-scale aerial image retrieval method based on the scale invariant feature transform(SIFT)operator and the vocabulary tree.This method could quickly and accurately retrieve similar images from massive and unordered aerial images.Firstly,the SIFT descriptor was used to represent the image and then the vocabulary tree was built using SIFT descriptors in the image library through the hierarchical K-means algorithm.Finally,the term frequency-inverse document frequency method which calculated similarity in query images and images collection were further utilized to obtain the similar images.Experiments with large-scale aerial images showed the reliability,accuracy and actual performance of the proposed algorithm.