Real-time Indoor Scene 3D Reconstruction Based On Kinect

With the rapid development of science and technology,3D reconstruction technology which is an important branch of computer vision technology has got more and more attention in the last few years.The application of 3D reconstruction involves more and more fields,such as video entertainment,archaeological exploration,everyday life and industrial production.Especially in 2010,Microsoft launched Kinect somatosensory peripherals with a depth sensor which can easily obtain depth information.Compared with the traditional 3D reconstruction method using monocular or binocular camera,the method using depth sensor has great advantages.The low prices of Kinect make it possible for ordinary people to use the technique of 3D reconstruction.In 2011,the real-time reconstruction algorithm called Kinect Fusion was proposed by Microsoft,it is the first time to realize the real-time 3D reconstruction based on the low-cost consumer camera.Although the algorithm can achieve good real-time performance,the size of scenes which needs to be reconstructed is limited.And the Kinect Fusion algorithm only uses the information of depth,so it can only produce a three-dimensional model without color texture.Therefore,this paper proposes a real-time 3D reconstruction method which can be used for large indoor scene based on Kinect Fusion and the analysis of point cloud registration method on the basis of ICP algorithm.This paper also presents a real-time 3D reconstruction system for indoor scene based on Kinect.The main research of this paper is as follows:1.In the preprocessing of depth image.First of all,the paper analyzes the reasons that the depth image produces noise,compares the de-noising effect of median filtering,gauss filtering,joint bilateral filtering,and bilateral filtering,and finally chooses the bilateral filtering as the de-noising algorithm.Secondly,the calibration of the Kinect camera is completed,and the parameters of the camera are obtained.2.In the step of point cloud registration.A detailed investigation on the current situation of point cloud registration technology has been made.Through analyzing the various steps of the ICP registration algorithm,especially the different methods of the resampling,corresponding points matching and the choice of the objective functions,this paper proposes a scheme which can be applied to real-time point cloud registration.3.In the step of point cloud fusion and surface reconstruction.Considering the requirement that real-time 3D reconstruction needs to fuse point cloud fast,this paper uses the point cloud fusion algorithm based on TSDF.The registration method of "frame-to-model" based on the global data cube is compared with the general "frame-to-frame" registration method.According to the experimental data,the former which has smaller accumulated error is used for the registration of the point cloud.Finally,the method of surface reconstruction based on Marching Cubes and the color images acquired by Kinect are used to get the model with color texture.4.Combined with the PCL,this paper presents a real-time indoor scene 3D reconstruction system based on Kinect which supports the Windows platform.The feasibility of the system is proved by the running results and experimental analysis.