Research On Scene 3D Reconstruction Algorithm Based On Line Features

How to obtain the 3D information of the scene environment through images has always been one of the hot spots in the field of computer vision.Image-based 3D reconstruction,as a technology to reconstruct 3D information of scenes from 2D images,has received extensive attention from scholars.At present,most of the image-based 3D reconstruction is based on point features.The number of point cloud models generated is huge,and there is almost no geometric topology relationship between points.It can not express the 3D structure information of the scene,nor is it conducive to the subsequent storage and calculation of the computer.At the same time,in the scene with weak texture,the number of point features is limited,which is easy to match errors,resulting in unsatisfactory reconstruction effect.Compared with point features,line features are rich in information and less difficult to match in weak texture scenes and other scenes.At the same time,line features are easier to describe the 3D structure of the scene.Usually,only thousands of line segments are needed to reconstruct the 3D model,which improves the efficiency of reconstruction.Therefore,this paper studies the scene 3D reconstruction algorithm based on line features.The contents are as follows:Firstly,this paper studies the line feature extraction method based on edge pixels.This method has the advantages of short operation time,high line segment continuity and strong anti noise ability.At the same time,the method of scale space and graph matching is combined to realize the matching of 2D line features.The matching method has high matching accuracy and strong robustness to image changes.Secondly,this paper introduces the COLMAP algorithm to obtain the camera parameter matrix,obtains the 3D line segment information through the pure geometric constraints and the clustering algorithm of 3D line segments,then uses the beam adjustment optimization model to extract the 3D line segments of the scene in a short time,which reduces the amount of data of the 3D model and improves the efficiency of reconstruction.Thirdly,based on the geometric relationship between line and line,line and surface,this paper proposes a RANSAC plane detection algorithm to detect multiple planes from 3D line segments.Then,by constructing the distribution equation,this paper reconstructs the smooth and watertight model surface for the cell distribution state cut by plane.