Scalable Platform For Real-Time Multi-Projector Augmented Reality

Because of the continuous improvement of performance and the shrinking volume, augmented-reality technology based on projectors has been widely concerned. Multi-projector augmented-reality system combines the virtual information with reality, projects visual information on complex surfaces of real object.There are still some problems that restrict the system in our laboratory. (1)Algorithms like geometric correction, photometric compensation, need to transform images by calculate every pixel, and there are large amount of calculation. So, most existing systems have a low real time property, can't give a fluent visual experience. (2) Existing systems need to use different constructs and image processing sequence for different screens. Degree of coupling of the system is very high, once the screen has been changed, we must redeploy all the system.Aimed at two defects above, this paper starts at algorithm structure and processing flow, improves existing system by modularization and distributed computing. The main tasks of this paper are:(1)We proposed the idea of modular to solve the problem of low scalability: various AR algorithms will be abstracted to processing units with the same I-O (input and output) structure, and then we connected different modules to a pipeline of image data. Once the original image data flowed through the units in the pipeline, the transformation is completed. We used scripts to configure the system, and scripts can be coded on the server to manage all the computers in the system. So, system deployment can be simplified as building blocks, scalability of the system has been greatly enhanced.(2)We focused on improving the real-time property of multi-projector augment-reality system. We found that the bottlenecks of the real-time system are computing ability and network bandwidth. In order to achieve high real-time property using general computers, we implemented a distributed rendering system, broke the bottleneck of computing ability. Then, we optimized the network and sorting method to reduce constrains of real-time property form network.(3)At last, we used graphic devices to calculate parts of algorithms, and achieved higher real-time property.The innovation of this paper mainly lies on these fields. We combined projector based AR and distributed rendering together, broke bottleneck of abilities in computing, improved the real time property. At the same time, we implemented our system with the concept of module, making the system having a feature of high coupling and low cohesion, improved the scalability. At last, we used graphic device to calculate parts of algorithms, increased the frame rate of the system.In order to project images on a new screen, we can add, modify or delete function modules to adjust the structure and flow. After distributing the computation and GPU acceleration, the frame rate can be greatly elevated, images can be projected smoothly, and meets the requirements for FPS of a AR system.