Research Of Three-Dimensional Scanning System Based On Kinect Sensor

Three-dimensional scanning technology is to build a three-dimensional model of the real world entity by utilizing digitization approaches in the computer. In the field of computer vision, reverse engineering, virtual reality and other research areas, three-dimensional scanning technology has been a hotspot and difficulty of the research and has a broad application prospect. At the same time, three-dimensional scanning technology is also one of the major challenges facing in the basic research and applied research.The traditional three-dimensional scanning equipment is often expensive and the operation is very complicated. Taking Kinect sensor which has low cost and simple operation as a data acquisition device, the three-dimensional scanning system studied in this paper can obtain multi-view depth data of the target and convert depth data into three-dimensional point clouds. After registration and integration of point clouds in different perspectives, a global three-dimensional model of the target can be acquired. In this paper, the main research work includes:First, we capture depth images of a target at different viewing angles by Kinect sensor and convert the depth information into three-dimensional point cloud data containing X, Y, Z coordinate information according to the transformation relationship of related coordinate system. Then we simplify the number of point cloud data in order to reduce the subsequent calculation.Secondly, we analyze and study the principles of point cloud registration as well as existing registration methods, and present an improved algorithm for the automatic point cloud registration combining with the characteristics of three-dimensional point clouds by the transformation of depth images collected by Kinect. We can get global three-dimensional data after registration through integration of all the point cloud data. The registration algorithm is divided into two stages which are the initial registration and fine registration. In the stage of initial registration, we need to estimate feature descriptors of the two point cloud data for registration, and search initial corresponding points, then calculate the initial conversion relationship refer to corresponding points filtered and filtrated. In the phase of fine registration, the iterative closest point algorithm is used to do further registration and the final conversion relationship is obtained on the basis of initial registration. Experiments show that this registration method can not only ensure the quality of point cloud registration, but reduce the computation complexity, and achieve higher maneuverability and better robustness.And then, in order to meet the visual requirements of a target, we use greed projection triangular mesh algorithm to reconstruct the surface model from three dimensional point cloud data and gain the three dimensional model of entity.Finally, three-dimensional scanning system software is programmed in accordance with the algorithm proposed in this paper. The feasibility of the system software is verified using the data of object obtained by myself.In this paper, the three-dimensional scanning system is easy to operate and low-cost without high performance computer configuration. Since Kinect itself is not very high in accuracy, the details of three-dimensional model obtained in this system is not accurate enough, but the general outline is clear and can meet the basic requirements for the application of imprecise computation.