3D Scan Reconstruction System Based On Gpu-accelerated

Computer3D reconstruction is an important area of research in computer-aided design and computer graphics. According to two or more2D images captured by computer, then the system obtain the two-dimensional geometry and topology of object, and then build three-dimensional model. The traditional three-dimensional modeling is generally used to obtain scanned data to modify and beautify.The data used for three-dimensional reconstruction generally come from the three-dimensional scanner. In traditional processes, the three-dimensional scanner capture the geometrical information, the most primitive information is the coordinates of the point data of the surface. Then these data is handled to the modeling software to be meshed to achieve three-dimensional reconstruction. During the process, due to the mechanical platform, area array CCD, laser generator, signal interference, and the measured object surface features such as hardware factors, measurement error, missing data phenomenon are very prone to be happened. Moreover, the scanning operation is relatively a slow procedure, and the user can not interactively observe the scan results. The purpose of this paper is integrating three-dimensional reconstruction process dynamically into the3D scan process, then the user can observer the process of object being scanned real-time.There is a lot of literature about three-dimensional reconstruction at home and abroad. Although they can get good quality of reconstruction, but most of them are used for a dedicated graphics workstation, expensive hardware costs limit the universal application of this technology. For ordinary small businesses, existing hardware cannot achieve real-time scanning point cloud meshing.With the development of the GPU, GPU’s parallel data processing capability has been greatly exceeded with the CPU of the same price. In light of this, this paper propose a parallel algorithm of three-dimensional reconstruction based on GPU, and applied to the three-dimensional scanner. It implemented that the user can obverse the meshing process of the scanned data real-time. So the user can notice the scanner error in time according to the mesh of scanned data. Achieve these do not need to increase the hardware cost, and this will greatly popularized the applications of three-dimensional scanner and three-dimensional reconstruction. Main tasks are as follows:Left and right CCD complement matching is the key issues in the3D laser scanning measurement system. This paper analyzes the error phenomenon and genesis of left and right CCD complement mismatching in the process of laser scanning measurement, and puts forward a new algorithm of left and right CCD image complement matching which is reconstructed3D depth information form two laser line images, the algorithm is based on the state information of five data points which in the same laser center line. The experiment showed that this algorithm improves left and right CCD complement matching accuracy.In the data processing stage, this paper proposed an algorithm based on GPU to quickly create octree, it has two novel features. First, it takes full advantage of the fine-grained parallel nature of GPU to build the octree in realtime. Unlike traditional methods, the CPU’s algorithm create tree by a depth-first traversal method, our technology use the level-order traversal. Process all the octree nodes at the same depth in parallel at one time. Today’s GPU structure contains a number of multiple physical processors, which can concurrent thousands of threads to fully exploit its performance. With level-order traversals, our technique maximizes the parallelism by spawning a new thread for every node at the same tree level.In this paper, a scanning reconstruction system based on GPU is designed and implemented. Research of the algorithm in the practical application of user have been tested, and meet the needs of users.