The 3D Point Cloud Reconstruction Of The Airship Impact Sequence

With the urgent needs of the digital three-dimensional modeling,the 3D reconstruction is also more and more important.At present,3D reconstruction is an active research direction,the direction of the research related to the computer graphics,computer vision and other disciplines Photogrammetry and Remote Sensing.Due to the interdisciplinary,which makes 3D reconstruction encounter the many difficultiesAbout 3D reconstruction,from getting data,is based on the ground data and the aerial data,or the intersection of the ground data and the aerial data.From the data source,the 3D reconstruction is based on image sequences and lidar data,or mixed those data.In this paper,we discuss 3D cloud reconstruction methods based on UAV image sequences.There are two general 3D cloud reconstruction methods based on UAV image sequences.One is the photogrammetric method,the other one is computer vision multi-view geometry method.Since these two methods have advantages and disadvantages,using photogrammetric methods deal with the UAV image sequences to obtain 3D point cloud which get a lot of attention in the country,but this method has some drawbacks.Therefore,this article will use computer vision multi-view geometry method to deal with the UAV image sequences to obtain 3D point cloud.Since the 3D point cloud reconstruction method based on UAV image sequences involves a lot of detail algorithm.For example,camera calibration,image search,image matching,solving image position and attitude,bundle adjustment,dense matching algorithm.This paper limited scope of the chapter,we focuses on image matching,image posture recovery,dense point clouds reconstruction algorithms.In this process,the innovation of this paper is as follows:Firstly,in solving the problem of low attitude UAV image matching process,those images with large distortion,dip a large degree of uneven overlap characteristics,according to the characteristics,we use SURF matching algorithm and error removal algorithm to protect fewer errors,and carry out specific experiments,experiments show that the algorithm can get the expected results.Secondly,in recovering the image of the global position and orientation process,this article describes two algorithms recover image global position and attitude.In which the incremental motion estimation method for solving(Structure From Motion)is major algorithm,but the algorithm has some drawback.This paper introduces a batch-type algorithm,and derive a formula for the algorithm.Thirdly,about the dense point cloud generation,this paper describes the growth style dense point cloud generation algorithm,and make related experiments,summarizes the advantages and disadvantages of the algorithm,in dense stereo matching,this paper introduced a semi-global matching algorithm(SGM),and introduce the algorithm in detail.Fourthly,in this paper,we carry out specific experiments about two different ideas,to get valid conclusions,analysis of the causes of related issues arising from the experiment,and provides the basis for subsequent researchers.