Research On 3D Reconstruction Technology Of Farmland Scene Based On UAV

Agricultural production intelligent technology to enhance agricultural production efficiency has a very large of promoting the role.The high-precision digital map of farmland scene is the key to realize the intelligentization of plant protection,and the 3D Reconstruction technology of farmland scene is an important research to realize the digitalization of farmland map.The 3D map recovered by the 3D Reconstruction can be used for tasks such as autonomous navigation of plant protection drones,identification of obstacles,identification of crop growth,and grid surface reconstruction.This article with visual 3D Reconstruction research background,based on agricultural mapping drone captured by the high-altitude cropland datasets 3D dense point cloud reconstruction algorithm expanded research.First,visual SLAM technique image for pose estimation,followed by a use has a precise position and orientation of image the constructs binocular imaging model,and finally based on semi-global stereo matching technique estimating disparity map to achieve dense point cloud reconstruction.For structure from motion-based(Structure from Motion,SFM)of three-dimensional reconstruction algorithm there is a high time cost issue,this study to enhance the scene threedimensional reconstruction efficiency study started.Proposed visual SLAM algorithm and stereo binocular semi-global matching binding method,synchronization is complete farm scene in a 3D dense cloud point of reconstruction.This method avoids the pose estimation and disparity map estimation separation performed,reducing the intermediate latency.For the current visual SLAM algorithm to meet the real-time performance and the use of stability is relatively weak in the characteristic problems of visual SLAM algorithm of stability of the study launched and precision.In this study,based on the stability better SIFT feature as SLAM algorithm tracking features,such algorithms can be successfully in the non-video stream to farm scenes Air Sign data set operation;and and made image and RTK-GPS fusion of SLAM method,to complete the image the pose estimation,and other open source were precision arithmetic comparative analysis by the absolute mean error comparison found the present method has a better table now.For farmland scene of repeated texture and weak texture is rich,resulting in scenes recovery is not complete problem,this paper to improve the scene of dense point cloud reconstruction ability of a study.In the cost calculation link of the semi-global binocular stereo matching algorithm,the operation of enhancing and preserving the texture is added,and at the same time,the corresponding elimination scheme is proposed for the point cloud noise.The improved algorithm with other open algorithms are obtained dense cloud point map for comparative analysis,and found that the algorithm dense cloud point map noise more less,repeating textures,texture weak region recovery stronger obstacle restore the integrity better.