Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Foveated rendering algorithms using eye-tracking technology in virtual reality

Marianos Nikolaos-Xenofon

Full record

---

Summary

The wide-spread availability of consumer grade virtual reality head mounted displays, has transformed virtual reality to a commodity available for everyday use. Nowadays, nearly everything with a display can be used to immerse the user in a VR world. From Smart phones to game consoles, everything now has VR extensions such as the Samsung Gear VR and the PlayStation VR.All this constant evolution around the VR world demands constantly better and more detailed head mounted displays. With the increasing use of 4K-8K Ultra High Definition displays and the push towards higher pixel densities for head-mounted displays, the industry is pressured to meet market demands for intensive real-time rendering. Since the current processors cannot deal with the increased demands for excessive resolution on the head mounted displays, new techniques of rendering must be implemented.The human visual system is often assumed to be perfect despite limitations arising from a variety of different complexities and phenomena. Humans have two distinct vision systems: foveal and peripheral vision. Foveal vision is sharp and detailed, while peripheral vision lacks fidelity. This lack of fidelity in the peripheral vision system is what new techniques of rendering, the so called foveated rendering techniques, are trying to exploit. Perceptually lossless foveated rendering systems and methods, seek to increase rendering performance by lowering image quality in the periphery, while maintaining the user’s perception of full HD rendering.In this thesis, we are trying to gather insights on how beneficial may the adoption of these methods at a commercial level be, by implementing and evaluating our own foveated rendering approach. To do so, we are using a Head Mounted Display unit [the Nvis SX 111] and an eye-tracking device [the Arrington Viewpoint-EyeTracker]. The foveated rendering technique developed in this thesis renders at three different layers of resolution. Apart from the foveal layer, which surrounds the area that the user is looking at and has full HD resolution capabilities, and the peripheral layer, which contains everything that is in the user’s peripheral vision and renders at 40% of the full HD resolution, we implement another layer which functions as a transition layer between them. This last layer renders at 60% of the full HD resolution, and it was added so that the user doesn’t notice the massive difference in resolution at the border between the other two layers.In order to have a more accurate picture of the results and the functionality of the algorithm created, we conducted a number of experiments involving 19 students from the institution. The users were asked to enter a virtual world and complete a small game. While the users were immersed in the virtual environment, we were monitoring the performance of the algorithm. During these experiments, a 57% decrease in the number of pixels shaded was recorded. Most of these pixels belonged to the peripheral layer. This decrease leads to a maximum increase of 18.3% regarding the number of rendered frames per second.This increase in FPS is the fundamental objective of this dissertation. Since we have achieved such an increase in FPS, it is safe to assume that foveated rendering algorithms are capable of large reductions in rendering cost using the latest technologies.

Available Files

Services