Real Time 3D Reconstruction System Of Large Indoor Scenes Based On Kinect

At present,the technical methods commonly used in scene 3D modeling are mainly based on two ideas.Forward thoughts and reverse thoughts.Forward thinking is based on 3D modeling software 3ds Max,Maya,CAD and other mainstream modeling software,but from idea to design to modeling,the time period is too long and the cost is high.Reverse thinking is based on traditional 3D reconstruction equipment,3D scanner,laser scanner,binocular stereo vision system,TOF depth camera.However,these 3D scanning devices are expensive,the structure of device is complicated,the amount of collected data is large,the scanning time is long,the real-time performance is poor,and the scene modeling effect is not ideal.With the advent and popularity of the consumer-grade RGB-D depth camera Kinect,3D modeling of large indoor scenes has become easier.However,Kinect’s depth information is limited by its own hardware,ambient lighting conditions,and distance.Its depth data range is limited and the size of the mesh model is fixed.It is usually only suitable for local and static indoor scenes.The depth data obtained by such depth camera scanning equipment is often noisy and voided,and the lack of geometric depth information makes 3D modeling difficult and inaccurate.In addition,the real-time 3D reconstruction of indoor scenes has problems such as cumulative errors and mismatches,poor dynamic real-time effects,limited scene reconstruction size,and no color texture information.In response to the above problems,this project intends to conduct research on key technologies for real-time 3D modeling of indoor scenes based on Kinect,and propose a low-cost,high-precision,real-time 3D modeling system for indoor scenes based on Kinect,a consumer depth camera,to solve the problem of depth data Key technical issues such as hole repair,high-precision multi-view depth data matching,real-time reconstruction and scene update.This article uses Kinect equipment to simplify and automate the 3D modeling of indoor scenes,reducing the threshold for 3D reconstruction.The work of this paper mainly includes:(1)Set up a low-cost,high-precision,real-time indoor large scene 3D modeling software and hardware system based on the consumer-grade depth camera Kinect.Scene information can be collected in an indoor environment,real-time 3D reconstruction and motion trajectory estimation of indoor scenes can be completed on the Linux operating system.(2)Using OPENNI to drive Kinect2 depth camera to obtain depth data and color data of indoor scenes.Accurately calibrate the Kinect camera to obtain the camera parameters,register and align the depth map and color map to generate a threedimensional color point cloud.This paper analyzes the reasons for the lack of largescale depth values in the depth data obtained by Kinect and the corresponding solutions,and compares several commonly used depth image filter repair algorithms such as Gaussian filtering,median filtering,mean filtering,and bilateral filtering.An improved depth map completion method is proposed,which studies the filling of holes in the depth map based on spatial correlation and color consistency constraints,and uses bilateral filtering to further repair and denoise.(3)In the Kinect-based real-time 3D reconstruction of large scenes,considering the real-time modeling of dynamic scenes,the ORB operator is selected to be used in the front-end image processing of the system.ICP algorithm has been optimized and improved.By simplifying and sampling the depth data,the quadtree structure is used to index the corresponding points,and combined with the ORB and RANSAC algorithms for feature detection and fast matching,which improves the efficiency of feature detection and registration.(4)With the increase of the 3D scanning time of the indoor scene,the accumulation of accumulated errors will cause mismatches such as overlap and misalignment to affect the overall effect of the reconstructed indoor scene.The research uses loop detection and global optimization to reduce the cumulative error,expand the Kinect Fusion algorithm to fix the scene size,and perform real-time 3D reconstruction and fusion rendering of large indoor scenes.