The Urban Scene 3D Reconstruction Algorithm Based On UAV Aerial Image

Recently,producing high precision 3D models from image sequence is a hot topics.High precision 3D models are of significant value in these fields,such as vehicle automatic steering,historical relic protection,digital city management.In urban scene,there are a lot of texture-less objects and wiry structures.Hence traditional point-based 3D reconstruction approaches may fail to integrate these structures into the resulting point cloud.In particular,some such as fences,wire tower scaffold or elongated structures,which are only a few points in the reconstruction point cloud.A different feature representation is necessary.Therefore this thesis propose a new three-dimensional reconstruction algorithm,which uses the estimated feature points and the additional line segments to be reconstructed.First,this thesis use UAVs to shoot urban scenes.Because small UAVs in the city can be a wide range of shooting,and freely from all angles on the target for the full range image acquisition.The resulting aerial image has large areas of overlap,the three-dimensional reconstruction provides a good image material.Then this thesis execute the traditional based on epipolar geometric constraints three-dimensional reconstruction algorithm to obtain pose by the camera.Recovery the initial structure of the initial position and orientation of the camera by the multi-view geometry theory.Utilize bundle adjustment algorithm that is based on set reprojection error constraints to unified adjust camera position and orientation and scene depth.Finally this thesis obtain accurate camera position and orientation.Later extraction and matching line segments.This thesis first use LSD(Line Segment Detector)line extraction algorithm for image segments extracted.To avoid matching the line feature based on the image itself,this thesis use pure geometric constraints in the line segment matching,and the use of camera pose match line segments for extracted line segment collection.It can tolerate a certain degree of shielding segments and has strong robustness.Finally,according to epipolar geometric constraint this thesis carry out line segment-based three-dimensional reconstruction algorithm.Based on the thought of clustering,the different images corresponding to the same three-dimensional line segment clustering,and then this thesis propose an incremental line segment-based three-dimensional reconstruction algorithm.By defining dynamic grouping radius,this thesis group the newly added image segments.After completing the match with an image,this thesis perform hypothesis verification and exclude unpromising line segments.At last,this thesis obtain the effective collection of three-dimensional line,completing the line reconstruction.A lot of experiments show that the proposed method can perform urban scene three-dimensional reconstruction effectively,dense sparse point cloud,build a linear structure.This thesis can gain valuable segment of three-dimensional model in the rich city scene.This algorithm prevents poor performance problems caused by large amounts of data,and can be efficiently complete the task of reconstruction.At the same time our algorithm is to gradually produce results,and lay a good foundation for future online three-dimensional reconstruction.