Research On 3D Reconstruction Of Large Scale Scene Based On Monocular Camera

As the focus and difficulty of computer vision,surveying and mapping,and other fields,3D reconstruction technology has always been the focus of research by stereoscopic vision researchers.The three-dimensional reconstruction of the real scene is a hot issue in the field of virtual reality and reality enhancement,and the three-dimensional reconstruction technology has a good application prospect in aspects such as medical diagnosis,reverse engineering,cultural relic protection,terrain survey and the like.In the three-dimensional reconstruction technology,image-based three-dimensional reconstruction is a cheap,convenient and easy to generalize a three-dimensional reconstruction method.Image-based three-dimensional reconstruction is to reconstruct a three-dimensional model of a target through a set of images taken from the target at different viewing angles.Using a monocular camera to reconstruct a device that is more economical and convenient for image acquisition than a professional 3D scanner.This article is based on the monocular camera images to study the algorithm of 3D reconstruction of large-scale scenes.It analyzes and studies the static 3D reconstruction of various scenes and dynamic targets,and designs 3D reconstructions under various situations reasonable algorithm.The main research work is as follows:1 Explain how to complete the three-dimensional reconstruction of the scene by means of camera calibration,image feature extraction and matching,point cloud reconstruction,point cloud triangulation,and texture mapping.The three-dimensional reconstruction of the image data obtained by the calibrated monocular camera can be performed,and a large-scale three-dimensional scene model with realistic visual effects can be obtained to verify the feasibility of this series of methods.2 Introduce a grid motion statistical matching algorithm that is different from the SIFT feature matching algorithm used in traditional 3D reconstruction,and reasonably utilizes the advantages such as high matching accuracy and large number of matches to solve the problem of current dynamic target 3D reconstruction difficultly.3 Analyze the principle and feasibility of restoring three-dimensional structure through relative displacement,and apply it to the three-dimensional reconstruction of dynamic objects to verify the correctness of the results and the feasibility of the algorithm.4 Propose a reasonable reconstruction method for dynamic target reconstruction in different situations.There are two main dynamic target reconstruction strategies:one is based on manual image segmentation of the target and the other is through automatic background segmentation and reconstruction of dynamic target extraction method with target tracking.