Research And Implementation On Multi-view Reconstruction And Point Cloud Contour Feature Extraction Technology

3D reconstruction technology can reappear rael scenes in computer,thus helping people understand environment information and object information more comprehensively.Unmanned aerial vehicles(UAV)which have high perspective and large field are widely used in urban planning,land survey and other areas.Aiming at the application of using UAV to evaluate the building situation afer disaster,this paper mainly studied multi-view 3D reconstruction technology and point cloud contour feature extraction technology,which are the two key technologies of thie application scenario.The specific research contents are as follows:First of all,this paper studies the relevant theories of reconstructing the 3D model from images,and calibrates the camera parameters which are nesssary for reconstruction,in order to establish the foundation of multi-view reconstruction.What is more,this paper extracts and matches the features of multi-view images,finds the appropriate initial image pairs,computes the external parameters,and gets an initial 3D model.After that,this paper selects one new image each time,computes the camera matrix through 2D-3D connection,recovers the 3D information,makes a global optimization,gets the sparse 3D model,and then uses dense reconstruction technology to further improve the model contour and detail information.Aiming at the need of breakpoint reconstruction and model improvement in practical application,this paper proposed a model-based view reconstruction method,which can add new views to the reconstructed model at any time to improve the model quality.Furthermore,this paper pre-processes the point cloud,uses segmentation and projection to get the point cloud plane,and extracts the plane contour feature to fit the line model.This paper introduces the contour connectability judgment before line fitting,which can divide the plane contour into a set of connectable contours,so as to solve the incorrect fitting problem when both inner and outer contours exist at the same time.Finally,this paper designs and implements a 3D reconstruction-contour extraction system,which includes five parts: date input,date output,visualization interaction,3D reconstruction and point cloud processing.It can reconstruct 3D model based on images and process the 3D model.It also supports 3D model export and line model parameters export,which is convenient for practical application.