Research On VR Rendering Optimization Algorithm Driven By Gaze Point Data

With the complexity of virtual reality application scenarios and the continuous improvement of hardware specifications of head-mounted displays,the number of rendering primitives and pixels that need to be processed by the GPU per unit of time have increased significantly.In recent years,researchers have applied the research results of the human visual system to the traditional real-time rendering framework,providing a new approach for rendering optimization.Unlike presenting a consistent high-quality image across the entire field of view,based on the difference in visual acuity of the human eye within the field of view,researchers can render a high-quality image in high visual acuity area and degrade rendering quality in low visual acuity area.Thereby,the real-time rendering performance of virtual reality is significantly improved.This rendering optimization algorithm driven by gaze point data is called foveated rendering technology,this thesis is based on this topic to carry out analysis and research.This thesis first describes the characteristics of the human visual system.Due to the uneven distribution of photoreceptor cells on the retina,the ability of the human visual system to obtain visual information in the entire field of view is not consistent.Imaging aliasing caused by different eye movement behaviors will further limit the ability of the human visual system to obtain visual information.This provides a theoretical basis for the method of improving rendering performance by reducing the image quality of peripheral areas in virtual reality.Then,for the visual acuity model commonly used in the existing foveated rendering technology,which only considers static visual acuity,This thesis analyzes and proposes a spatial frequency model,and uses this model to derive the minimum resolution size of the human eye at any position in screen space.The spatial frequency model considers both static visual acuity and dynamic visual acuity.It divides the rendering space into three rendering areas,and each rendering area can use different rendering strategies according to the difference in the minimum resolution size of the human eye.to render the final screen image through fusion.Finally,the final screen image is rendered through fusion.Finally,to solve the problem that the existing foveated rendering technology usually only optimizes a single rendering space,This thesis proposes a hybrid foveated rendering technology driven by gaze point data.The hybrid foveated rendering technology combines the spatial frequency model to optimize the rendering of the two rendering spaces,the model space and the screen space,to further improving rendering performance.The experimental results show that the hybrid foveated rendering technology has higher rendering efficiency than the existing foveated rendering technology.