Research On 3D Reconstruction Of Indoor Scene Based On Binocular Vision

3D reconstruction is one of the hot research directions of computer vision.Indoor scene is one of the most important application scenes in 3D reconstruction,which has been widely studied.For example,indoor service robots recognition objects,human-computer interaction in VR,online display of cultural relics and online display of house,etc.The traditional 3D reconstruction method of indoor scene may not get the good target texture information,and it has the problems of high cost,low efficiency and heavy workload.To solve this problem,a 3D reconstruction algorithm for indoor scenes based on image feature points is proposed.In this paper,some complex areas image of the living room in indoor scenes are taken as the research object.This area usually has complex scene structure,rich color information and diverse reconstructed objects.By shooting images from different angles and using the reconstruction algorithm proposed in this paper,the 3D reconstruction of indoor scenes can be realized quickly.The main research contents are as follows:Firstly,this paper builds a simple and practical binocular camera system for the indoor scene,and selects the appropriate camera calibration algorithm through the analysis of the applied indoor scene.After the camera calibration completed,this paper use stereo rectification algorithms to improve 3D reconstruction accuracy.Secondly,the color information and light information are lost in the process of image graying.It will compensates all the gray images and more feature points will be collected.Thirdly,aiming at the contradiction between real-time and matching accuracy in feature point collection and matching,an improved Harris-SIFT algorithm is proposed to improve the realtime and matching accuracy in feature points collection and matching.Finally,this paper is supplied a new feature matching algorithm to 3D reconstruction,and an improved binocular structure from motion(SFM)algorithm is proposed to improve the real-time performance of SFM,so as to improve the real-time performance and accuracy of 3D reconstruction.Use the PMVS algorithm to densify the point cloud,and then use the normal vector information of the PMVS algorithm to reconstruct using the Poisson surface reconstruction algorithm.At last,the key points of the proposed 3d reconstruction algorithm for indoor scenes are verified experimentally.A binocular camera was used to shoot 12 pairs of images on the calibration plate for camera calibration experiment,and 30 pairs of binocular images were obtained through the shooting of indoor scenes for matching experiment and building 3D reconstruction experiment.Through experimental comparison and quantitative analysis,the effectiveness of the improved algorithm is verified.