Real-time Study Of Augmented Reality 3D Registration Algorithm Based On GPGPU

Augmented Reality (AR for short), also known as mixed reality, show great value in medical care, the monuments restoration, cultural heritage protection, industrial maintenance, virtual training, entertainment and games areas, more and more researchers are concerned about it because of its broad application and development prospects.3D registration is the key technology of augmented reality, it is the core to achieve the integration of virtual and real environment, which requires a higher real-time. However,3D registration algorithm based on the natural features seriously affect the real-time nature of the AR system due to the model complexity and huge amount of computation, the researchers improve the efficiency of the implementation of the algorithm for meeting real-time running needs by simplifying the algorithms or adding a secondary logo in the real scene. However these limit the application of AR system or even to produce registration error. Therefore, that needs further study about how to maximize the real-time AR system under ensuring enhanced information display quality.In recent years, GPU general computing become a hot topic of research at home and abroad, using the GPU to speed up the calculation can usually get an order of magnitude speed increasing, the GPU powerful computing and parallel processing capability is used to accelerate augmented reality 3D registration algorithm which is of great significance for building real-time and accurate augmented reality systems.The main research works are:(1) Firstly, research the GPU general computing and key technologies of augmented reality. Introduce the GPU general computing model and common technology. Then describe display technology, photorealistic rendering technology and 3D registration techniques involved in the augmented reality in detail. The focus is on augmented reality 3D registration algorithm and its key steps based on natural feature points.(2) Research the complexity of 3D registration algorithm based on the natural feature points and finds the speed bottleneck, parallel analysis about bottleneck part and use GPU to accelerate the achievement. There are two ways to accelerate bottleneck part using the traditional GPU general computing and based on the CUDA platform, the paper give detailed algorithm design and implementation process, at the same time, two implementations were compared with the traditional CPU implementation, experimental results show that the two methods have gained a certain speed on the upgrade. Finally, the paper introduces the specific implementation process of the AR system based on GPU and outlooks development prospect of augmented reality.