Research Of UAV Hyperspectral Images Stitching Based On 3D Reconstruction

Unmanned aerial vehicle(UAV)remote sensing has some irreplaceable advantages compared with traditional satellite remote sensing and manned aerial remote sensing.Traditional manned aerial remote sensing is subject to long flight time,dangerous shooting environment and bad weather,and satellite remote sensing may not be able to obtain full information in areas of interest due to bad weather and time.However,the UAV remote sensing due to its low cost,low risk and low flight attitude have become a supplementary means to obtain remote sensing information in addition to satellite remote sensing and aerial manned remote sensing.UAV remote sensing has good application prospects in the field of land resources survey,overall situation assessment of survey area and disaster emergency response,and the fast and accurate UAV images mosaic is one of the important data processing steps.Traditional UAV images stitching methods use Digital Elevation Model(DEM)to ortho-rectify images in the survey area,and then completes the stitching according to the geographic information of pixels.Although this kind of stitching methods can obtain high precision orthographic mosaic image,the stitching process is tedious and complicated,requiring manual intervention and the process take a long time.Therefore,an automatic mosaicking method based on 3D reconstruction is proposed in this paper.A dense point cloud is reconstructed by using the structure from motion(SFM)and multi-view stereo(MVS)algorithms,and then the coordinates of fixed points are calculated through some interpolation algorithms on the dense point cloud,after that,the indirect differential correction method is used to ortho-rectify the images,and the geometric consistent mosaic image is obtained by using the pixel geometric information.The experiment result shows that the proposed method has following advantages.First,there is no need of manual intervention in the whole stitching process.Second,the only input of the method is UAV images in the survey area,which neither requires the POS data at the camera shooting moment nor the control points data of the survey area.Third,it takes a short time to finish the mosaic by using the proposed method,the most time-consuming part of the whole process is feature extraction and feature matching,but by the GPU acceleration and using parallel processing techniques,the time of feature extraction and matching is shortened greatly,and the stitching speed is improved.