Sparse-reconstruction Of Aerial Image Of UAV Based On Incremental Structure From Motion

With the development of unmanned aerial vehicle(UAV)and digital camera technology,UAV aerial image processing has become one of the hot spots in current research.UAV aerial images have relatively large attitue angle and irregular flight course which will cause not to satisfy the processing requirements of classical aerial photogrammetry under the situation that accurate camera information cannot be obtained with uncalibrated camera.Incremental Structure from Motion method is the most common forms of Structure from Motion(SfM)in the field of computer vision.It estimates the pos of camera by detecting and matching features,selecting the initial image pair and estimating the pose,calculating the point cloud,selecting additional images,adjustment and other steps.The algorithm relies on accurate camera parameters with low degree of reliance,high degree of automation,and high precision.It can be used to recover the pos of a drone with large attitude angles,to make up for the shortcomings of traditional photogrammetry methods to some extent.However,there are still some problems in the classical incremental reconstruction method,which affects the efficiency and accuracy of reconstruction.Based on the classical incremental reconstruction method,this paper improves the image correspondence determination and the initial image pair selection,and realizes the restoration of the aerial image pose of UAV based on the improved algorithm.The main research contents and achievements of the paper are as follows:(1)The classic RANSAC algorithm and adaptive RANSAC algorithm are studied.The advantages and disadvantages of the two methods are analyzed and compared.The experimental results show that the adaptive RANSAC algorithm is superior to the classical adaptive RANSAC algorithm and can be used for Removing mis-matching with geometric constraint and pose recovery of intial image pair.(2)The image retrieval algorithm based on SIFT operator and vocabulary tree was introduced into the classical incremental reconstruction method to retrieve similar images.The image retrieval of five sets of data sets was used to verify the efficiency and accuracy of the algorithm.This retrieval algorithm is feasible in the introduction of classical incremental reconstruction.(3)Using the multi-model constraint method to describe three types of image pairs,it is verified that the multi-model constraint method can effectively describe the image pair model and provide the basis for the selection of initial image pairs in incremental reconstruction.(4)Improved incremental reconstruction method.Taking the aerial image of seven groups of drones as experimental data,the efficiency and accuracy of the improved method are analyzed.The experiment results show that the improved incremental reconstruction method improves the speed of image feature matching,effectively recovers the image pose so as to get sparse point cloud data,and largely improves the speed of sparse reconstruction.