基于 SIFT 算法的无人机遥感图像拼接技术

为了给农田研究人员提供高精度、 宽视野的图像, 在利用 SIFT(Scale-Invariant Feature Transform)算法初 步检测候选点步骤中, 加入自适应阈值去除部分候选特征点; 结合无人机图像的经纬度坐标及重叠区域位置关 系剔除部分无效特征点, 并进行特征点粗匹配; 利用随机采样一致算法消除误匹配点对, 并求解投影变换矩阵 完成相邻两幅农田遥感图像的拼接; 设计了金字塔拼接策略, 完成 128 幅高分辨率图像的拼接。 实验结果表 明, 基于 SIFT 算法, 利用改进的特征点精简方法, 特征点粗匹配时间平均减少了52%, 精匹配时间平均减少了 25%; 基于 6 个图像融合评价参数的对比实验发现, 从定性和定量两个方面, 基于多分辨率的图像融合均优于 其他融合算法。

Matching Technologies of UAV Remote Sensing Image Based on SIFT

In order to provide a panoramic image with high precision and wide field,the main purpose of this study is to achieve a high-resolution images stitching.For a large number of high-resolution farmland remote sensing images taken by UAV (Unmanned Aerial Vehicle),to obtain the full panoramic farmland image,the image mosaic algorithm is improved by combining with characteristics of UAV image.In detecting the candidate points step preliminarily using SIFT (Scale-Invariant Feature Transform) algorithm,we use adaptive threshold to remove part of the candidate feature points.By combining with latitude,longitude coordinates and the relative position relation of the overlapping area about UAV image,we remove part of invalid feature points and make rough matching on feature point.For completing two adjacent farmland remote sensing images stitching,we apply random sample consensus algorithm to eliminate mismatching point,and solve the projective transformation matrix.In order to complete 128 high-resolution images stitching,we design pyramid stitching strategy.The experimental results show that:with the improved feature points streamline method on SIFT algorithm,the time needed for the rough feature points matching is reduced by an average of 52％,and 25％ reduction for Accurate feature points matching.In comparison experiment based on six evaluation parameters,we found that the multi-resolution image fusion algorithm is superior to other fusion algorithms in qualitative and quantitative analysis.The study provides an efficient reference for a large number of high-resolution image stitching.