An Interactive Graphics Display Architecture

In recent times Virtual Reality has become a growing area within computer graphics with many and varied applications. Many systems suffer from inadequate performance resulting in a lack of realism and even psychological side effects[1]. A large body of research exists to improve the performance of Virtual Reality systems by increasing the rate at which scenes are drawn. This project differs in that the display controller is redesigned so as to cater directly for the needs of a Virtual Reality display system. By performing complex computation within the display controller as a pixel is being displayed, latency to user head rotations may be directly reduced and latency to translations and stereoscopy may be indirectly reduced. Simulations have shown this architecture to be feasible to implement with current off-the-shelf technology. A prototype is currently being constructed. 1.0 Introduction. This paper introduces and briefly investigates a new computer graphics display controller architecture which is suited to the requirements of head mounted Virtual Reality (VR) display systems. This architecture differs from standard display controllers in the nature of the video memory addressing mechanism. Rather than fetching pixels for display from the video memory sequentially, the fetch mechanism bases the pixel fetch location on the pixel's actual screen location (as seen through the wide angle viewing lenses) and the orientation of the user's head. This means the viewport mapping is delayed until after the scene has been rendered. The new display controller directly reduces latency to user head rotations by fixing user head rotation latency to the refresh period of the display device. Latency to other forms of interaction such as user translations through the virtual scene, head motion parallax and stereoscopy are indirectly reduced through the use of image overlaying. Image overlaying or image composition [2] is a technique often used by low-cost non-head mounted display graphics systems such as video games to increase the apparent realness of a scene. It is also often used by high performance systems to increase the maximum rendering rate. Different sections of the visible scene are drawn into separate display memories then overlayed to form a final scene (a similar technique is often used when creating cartoons to reduce the drawing required per display frame). When translating through the virtual scene, each display memory may be updated independently. Close objects receive more of the rendering engine’s time than less frequently changing background objects. Image overlaying generally does not work very well with conventional display controllers in VR systems as user head rotations invalidate all display memories including those containing background images. This mass invalidation occurs because all of the images in the display memories are rendered with a fixed viewport mapping so when the viewport changes due to a user head rotation all of the images have an incorrect viewport mapping. The new display controller does not suffer the same drawback since the rendered scenes in display memory are independent of the viewport mapping orientation and do not become invalid when the viewport mapping changes1. 2.0 A New Display Paradigm. Display generation usually consists of two processes. The first process called rendering draws images into display memory. The second process sends the image in the display memory to the display device. This process is performed by the display controller. Such a display system is shown in Figure 2.1. Database traversal Geometric transform. Trivial accept/reject