Research And Implementation Of Point Cloud Registration Technology Based On Kinect

The three-dimensional point cloud model shows the position,color,and shape of the object and presents three-dimensional graphics with a physical depth of field and vivid effect of display.So a realistic three-dimensional space is established at low cost,which solves the problems of traditional modeling software.It has a wide range of applications in the fields of cultural relics protection,telemedicine,and 3D games.This paper takes the point cloud data obtained by Kinect sensors as the research object and researches on the point cloud data registration and related technologies.We designed a point cloud registration system based on Kinect combining Kinect for windows SDK development kit,a set of C++ programming language and PCL point cloud library.The system consists of three modules based on functional requirements:point cloud data acquisition,point cloud preprocessing,and point cloud registration.In the point cloud acquisition module,we use the interface functions provided by Kinect for Windows SDK 2.0 to drive the Kinect device to obtain colored point cloud data.In the point cloud preprocessing module,the obtained point cloud data is used as the input,and the target point cloud is first extracted through the direct-pass filtering,and then the noise are processed through radius filtering combined with bilateral filtering.In the point cloud registration module,we take the preprocessed point cloud data as the input,and we use an improved initial registration method for extracting key points based on Hausdorff distance,through voxelized sampling,and then calculating the Hausdorff distance of each point and its neighbor points,drawing its normal distribution map,dividing the point cloud into key points Non-key points,feature matching and singular value decomposition of key points.After obtaining the initial value of registration,an optimized ICP algorithm of BBF optimized kdtree is used to complete the accurate registration of the two-frame interpoint cloud model.Finally,the multi-view alignment algorithm for point cloud sequences in different views is adopted to unify point clouds from different perspectives into the same coordinate system to complete the global registration of 3d point cloud data.After testing,the interface of the system shows the experimental results intuitively,and the system finally gets a point cloud model of the laundry detergent barrel after collection,preprocessing and registration.The cost of Kinect equipment is low,and the specific steps of the system are simple and easy,so the system has a certain practical value.