无人机影像匹配中尺度不变特征应用改进

An improved SIFT algorithm for UAV image matching

针对尺度不变特征变换算法用于无人机影像匹配中存在匹配效率低下,误匹配较多等问题,该文提出一种特征点提取优化算法,并改进了特征点匹配策略。通过将原始影像划分格网,依据每个格网影像信息熵的大小来合理分配各格网中特征点数量,实现了基于纹理信息丰富程度的特征点均匀分布;基于Harris兴趣值进行筛选,保留了适合于摄影测量的特征点;采用一种多层次自适应的匹配策略,在尽可能保留正确匹配的同时提高了匹配正确率。基于一对由小型无人机拍摄影像的实验结果表明,所提方法在大量减少SIFT特征点的同时,保证较高的正确率,增加了匹配点的精度,且提高了算法效率。

Aiming at the application of scale invariant feature transform （SIFT） algorithm in UAV image matching, this paper presented a new point extraction method and improved the matching strategy. The original image was divided into grid, and the uniform distribution of feature points based on texture information was achieved by properly allocating the keypoints based on image information entropy of each grid. The feature points suitable for photogramrnetry were reserved based on the Harris value of inter- est. A multi-level adaptive matching strategy was adopted to improve the matching accuracy while retaining a high correct matching. Experiments with UAV images showed that the proposed algorithm not only re- duced feature points, but also had greater right-matching ratio, which increased the matching accuracy and improved the efficiency.