一种面向无人机航拍图像的快速拼接算法

A Fast Stitching Algorithm for UAV Aerial Images

针对现有图像拼接方法对分辨率高、影像信息复杂的无人机航拍图像处理速度慢、匹配精度差的问题,提出一种面向无人机航拍图像的快速拼接算法。首先,利用FM(Fourier-Mellin)算法求出图像的重叠区域,并获取图像重叠区域内的感兴趣区域,结合区域分块以及图像信息熵得到最终的特征提取区域,设置为图像掩膜;接着,在特征提取区域提取特征点,并且对特征点匹配对进行两轮筛选,减少RANSAC算法迭代次数;最后用改进的RANSAC算法进行误匹配点对的剔除以筛选出准确度较高的匹配点对。比较该算法与其他算法的运行效率以及图像的拼接质量等相关指标。实验结果显示该算法较SURF、SIFT、区域分块算法,航拍图像拼接时间分别降低35%、56%、57%,满足航拍图像对拼接精度以及实时性的要求。

Aiming at the problems of slow processing speed and poor matching accuracy of the existing image stitching methods for UAV aerial images with high resolution and complex image information,a fast stitching algorithm for UAV aerial images is proposed.Firstly,the FM(Fourier-Mellin)algorithm is used to find out the overlapping region of the image,and obtain ROI from the overlap area of image,combining the area blocking and the image entropy to get the region of final feature extraction as the image mask.Then,the feature points are extracted in the area of feature extraction and two rounds of filtering are performed on the feature matching pairs to reduce the number of RANSAC algorithm iterations.Finally,an improved RANSAC algorithm is used to reject the wrong pairs in order to filter out the pairs with higher accuracy.Compared the operation efficiency of this algorithm with other algorithms,as well as the image stitching quality and other related indexes,the experimental results show that this algorithm reduces the stitching time by 35%,56%and 57%compared with SURF,SIFT and area blocking algorithms respectively,meeting the requirements of today’s aerial images for stitching accuracy and real-time performance.