Detailed 3D Modeling Of Building Based On UAV

With the rapid development of drones and computer technology,high-quality and high-precision 3D modeling technology for buildings has become a key research issue in fields such as cultural relics protection and smart cities.Although the UAV tilt photogrammetry model can objectively and realistically reflect real-life buildings,due to the incomplete acquisition of some images near the ground and occlusion,there are still deficiencies and defects in the fineness and integrity of the building model elevation.The UAV image has the characteristics of large differences in perspective and scale,and there is a problem of low image matching accuracy,which leads to the poor quality of the reconstructed building 3D model and cannot meet the needs of refinement.Therefore,this article focuses on the problems in the refined 3D reconstruction of buildings,using a modeling method from coarse to fine.Based on the initial point cloud model of the building,plan a close to photography route,and studies the UAV image matching and point cloud model registration in the 3D reconstruction process.Through the registration and fusion of the initial point cloud model of the building and the facade point cloud model,the facade information of the initial point cloud of the building is filled to realize the fine 3D modeling of the building.The main research contents are as follows:(1)Aiming at the problem that the lack of building facade information collected by oblique photogrammetry leads to the distortion of the model facade texture,a method of planning UAV close to photography route based on the initial point cloud model of the building is proposed to supplement the building facade information.Firstly,the single-lens UAV is used to collect the initial information at the top of the building,and the initial model of the building is used to guide the planning of the elevation close to the photography route,the viewpoint position of the UAV covering the building facade is calculated according to the flight height,overlap degree and the visual field size of the UAV.The particle swarm optimization algorithm is used to plan the optimal aerial photography path,complete the high-resolution building facade image acquisition.(2)Aiming at the problem of large viewing angle and geometric distortion in UAV sequence images,it is difficult to obtain reliable feature points,an improved affine scale invariant feature transform algorithm based on UAV position and orientation system is proposed,which improves the stability and matching efficiency of feature points.Due to the low matching accuracy caused by many mismatches,the mismatch elimination based on the nearest neighbor distance ratio method and the progressive sample consensus algorithm is proposed to complete the accurate matching of UAV images.Experiments show that the method extracts more feature points,and the algorithm efficiency and matching accuracy are higher.(3)Aiming at the problem of low registration accuracy and poor stability of building initial point cloud model and facade point cloud model,a registration method based on the initial registration of the intrinsic shape signatures and the 3D shape context feature and the fine registration of the improved iterative closest point is proposed.Firstly,feature points with more uniform distribution are extracted based on intrinsic shape signatures and 3D shape context,and coarse registration is completed based on sample consensus initial alignment algorithm.In order to further reduce the registration error,the iterative closest point algorithm based on normal vector constraint is used to realize the precise registration of the initial and facade point clouds of the building.The implementation shows that this method can achieve high registration accuracy.Taking a single building in the campus as an example,based on the image data collected by a single-lens UAV,the accurate matching of UAV images and the registration and fusion of the initial and facade point cloud models of the building are completed to obtain a complete building model.Comparing and analyzing the building tilt photography model and the model built by this method from two aspects of model accuracy and texture structure,the model built by this method has higher accuracy and clear detail texture,which realizes the fine modeling of buildings and illustrates the effectiveness and feasibility of this method.