The 3D Reconstruction And Recognition Based On Kinect

The technology of 3D recognition and reconstruction can reveal the object of the real world in the form of the computer information by the means of digital computer.In recent years,The technology of 3D recognition and reconstruction has obtained the rapid development in many fields,for example medical research, cultural relics restoration, recognition technology,instrument manufacturing,game projects research.The heart of the problem is how to get the 3D information of any scenes or objects in different perspectives and scale,and to exact fuse and identify the form of any surfaces of scenes or objects by a real-time and high-efficiency method. Traditional methods of 3D recognition and reconstruction includes structured Light 3D scanning,laser scanning and so on.However,These methods have problems of equipment complicated,price highly and Operation trival complex.We present a scanning instrument which could get the depth information from the real world calld Kinect. The objects from the real world would be handle through the following process : surface measurement, surface reconstruction update, surface prediction and sensor pose estimation.Then,we would research point clouds got from above process and build range images and scale space,calculate suface curvature,identify the form of the shape index and curvedness. The main content of this thesis as follows:(1) First of all,we introduce the structure, using principle and technical characteristics of Kinect,especial in the hardware structure.Here,we emphasis on the core chip of Kinect. In addition to,we describes the principle of system calibration,especial in camera calibration, coordinates calibration.In the end,we introduce some filter methods in the process of the point cloud pretreatment.(2) We present the method of getting point cloud and the process of it when use the Kinect to obtain 3D reconstruction.In this part, the algorithm of getting and reestablishing complex surfaces would be introduced,which include the process of obtaining the intensive vertex of the depth image,predicting the surface of objects and evaluating the pose of camera. In addition to,another topic is the splice of point cloud which include initial splice and ICP splice.(3) Describe the algorithm of surface form identification.Firstly,we introduce the concept of range image,In this process,the image feature should be remain unchanged when the camera position rotating and mobile shooting by the method of keeping the image orientation invariance.Secondly,By calculating the principal curvatures, Gaussian curvature, mean curvature, we could get Shape index and curvedness of the surface,which could fitting surface and determine the surface type.