Gaze-Contingent Level Of Detail Rendering

The contributions of this paper are the development and evaluation of a nonisotropic model-based Level Of Detail (LOD) rendering technique for gaze-contingent viewing of multiresolution meshes. A high resolution portion of the model is rendered at the eye-tracked Point Of Regard (POR). A method is given for converting a closed polygonal mesh to a nonisotropic LOD representation suitable for gaze-contingent viewing. Based on a theoretical model of visual acuity, a three-dimensional spatial degradation function is obtained from human subject experiments in an attempt to render imperceptibly degraded geometric objects. Unlike previous LOD approaches, our resolution degradation method is based on the measurement of visual angle in world coordinates and is applied directly to object geometries prior to rendering. The gaze-contingent technique is evaluated in a Virtual Reality (VR) system integrated with a binocular eye tracker. To our knowledge, this is the first example of a binocular eye-tracked VR system used to evaluate a gaze-contingent modeling technique. Results are reported in terms of rendering performance, indicating an overall 4-fold average frame rate improvement during gaze-contingent viewing. Frame rate improvement ranged from a factor of at least 2, up to a 15-fold gain in performance over full resolution display, varying with the model complexity and the instantaneous direction of the viewer’s gaze.