The Research On Point Cloud Registration Technology Based On Depth Camera

3D reconstruction is a hot topic involving many fields such as computer vision,virtual reality,reverse engineering,robot autonomous navigation and so on.On the one hand,with the development of depth camera technology,the depth camera represented by Kinect is more and more widely applied to 3D reconstruction due to its advantages of simple operation,low price and good performance.On the other hand,all the data needed for reconstruction can not usually be obtained at one time due to the influence of object shape,environment and equipment limitations.Therefore,point cloud splicing is the focus of 3D reconstruction.Data collected from different angles has different coordinate systems.Point cloud registration means that the point clouds under different coordinate systems are unified into the same coordinate system according to an algorithm to obtain the complete model data of the object to be reconstructed.In this paper,Kinect is used as a data acquisition device to study the theory and experiment of the generation and registration of point clouds.Then,main contents of this paper are as follows: Firstly,the working principle of Kinect camera,TOF,is described.And the PSO algorithm(Particle Swarm Optimization)was improved,the PSO algorithm is applied to camera calibration to get a more accurate conversion relationship between two-dimensional pixel coordinate system and three-dimensional world coordinate system.Secondly,the RGB image and the depth image are aligned.In addition,the newly introduced directional filtering algorithm and the FMM(Fast Marching Method)are introduced to filter the depth image.Then,the point cloud is generated according to the conversion relation obtained by the camera calibration.Last,the research of point cloud registration is carried out in this paper.Point cloud stitching is divided into initial stitching and precise stitching two parts.In this paper,the SAC-IA algorithm is used to initialize the point clouds,and the ICP algorithm is used to stitch the point clouds accurately.According to the process of point cloud registration,this paper proposes to use ORB-SVD method to provide initial iteration value for ICP algorithm and to optimize ICP stitching speed by downsampling,replacing the European distance by point-to-area distance and using KD tree search.Experiments show that the above improvements have achieved the desired results.