基于SIFT和TPS算法的四旋翼无人机图像自动配准

Automatic Registration of Quad-Rotor UAV Imagery based on SIFT and TPS Algorithm

针对四旋翼无人机图像姿态倾角大、图像变形明显等特点,采用尺度不变特征变换(SIFT)算法和薄板样条模型(TPS)对四旋翼无人机图像进行特征点匹配和配准实验研究,从拼接图像的目视效果和配准均方差方面比较分析了TPS模型与常用的仿射变换及多项式变换模型的图像配准效果。结果表明:在SIFT算法精确的同名点匹配下,TPS变换模型能够兼顾四旋翼无人机图像的整体刚性变形及局部的非刚性变形,无论是目视效果还是均方差定量分析,TPS变换的图像配准精度最高、效果最好,能够满足四旋翼无人机图像的快速配准、拼接要求。

As an important complementarity of remote sensing technique,Unmanned Aerial Vehicle （UAV） is at tractive to various applications such as target searching, island management, disaster monitoring and low-altitude photogrammetry. The qua＆rotor UAV, Compared to fixe＆wing UAV, has the advantages of Vertical-Taking-of Landing （VTOL） and low-altitude flexible flight. However, the qua＆rotor UAV has small-volume and light weight and is more easily influenced by the wind, the acquired imagery maybe have bigger tile angle and more obvi- ous geometry distortion will would result in more difficulty on the image feature matching and image mosaic. For this question,we apply the Scale Invariant Feature Transformation （SIFT） to image feature matching and Thin Plate Spline （TPS） transformation to automatic registration of the micro qua＆rotor UAV imageries in this paper. The registration imagery based on TPS transformation is then compared to that based on the affine transformation and the polynomial transformation by evaluating the visual effect of the mosaic imagery and Root Mean Square （RMS） statistic, which shows that after precise SIFT feature matching, the registration RMS accuracy and the visu al effect of mosaic imagery of TPS transformation are best, and satisfies the demand of rapid registration and mosaic of micro qua＆rotor UVA imagery,since TPS transformation considered both rigid transformation and partial non linear distortion of the micro qua＆rotor UAV imagery.