作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• •    

光学透视式AR显示中环境光驱动的自适应注视点渲染方法

  • 发布日期:2026-11-20

An Ambient-Light-Driven Adaptive Foveated Rendering Method for Optical See-Through AR Displays

  • Published:2026-11-20

摘要: 针对光学透视式增强现实显示中虚拟内容易受环境光冲淡,以及移动平台算力和显存带宽受限导致渲染效率不足的问题,提出一种环境光驱动的自适应注视点渲染方法。该方法将环境照度映射为环境光耦合系数,联合调节高分辨率注视区域和低分辨率背景缩放比,使注视区大小与外围分辨率随环境光强动态变化;环境光越强,注视区域越小、外围分辨率越低,以降低片元着色与显存带宽负载;环境光越弱,则适当扩大注视区域并提高外围分辨率,以保障画质。方法采用子视锥体投影矩阵裁剪局部渲染范围,从几何投影阶段减少传统视口裁剪造成的边缘错配与拉伸伪影;将背景采样、注视区域融合和环境光混合合并为单次全屏绘制,以降低中间纹理读写与全屏拷贝开销。本文聚焦渲染算法层面,实验采用离线眼动回放与仿真环境光以控制变量,重点考察算法对环境光照与注视轨迹输入的响应。实验基于公开眼动数据集在多场景下进行,在复杂场景强光条件下,该方法相较固定注视点渲染和全分辨率渲染的帧率最高分别提高35.73%和54.71%;除渐变光照、随机扫视的复杂场景外,其余动态测试配置的帧率变异系数均低于8%。同时,环境光的冲淡效应有效掩盖了因外围分辨率降低而导致的客观图像质量损失,使本方法的主观评分与固定注视点渲染基本持平,强光下差异更小。消融实验表明,子视锥体投影、环境光自适应参数决策和单次全屏合成均具有独立正向贡献,并存在协同增益。性能实验结果表明,环境光冲淡效应可作为外围分辨率分配的辅助感知依据;在算法设计层面,能够为资源受限的光学透视式增强现实的渲染优化提供理论支撑。

Abstract: To address the problems that virtual content in optical see-through augmented reality (OST-AR) displays is easily washed out by ambient light and that rendering efficiency is limited by constrained computing power and memory bandwidth on mobile platforms, an ambient-light-driven adaptive foveated rendering method is proposed. The method maps ambient illuminance to an ambient light coupling coefficient, jointly adjusting the proportion of the high-resolution foveal region and the scaling ratio of the low-resolution peripheral background, so that the foveal region size and peripheral resolution vary dynamically with ambient light intensity: the stronger the ambient light, the smaller the foveal region and the lower the peripheral resolution, thereby reducing fragment shading and memory bandwidth loads; the weaker the ambient light, the larger the foveal region and the higher the peripheral resolution, thereby preserving image quality. A sub-frustum projection matrix is used to clip the local rendering range, reducing edge mismatch and stretching artifacts caused by conventional viewport clipping at the geometric projection stage. Background sampling, foveal region blending, and ambient light composition are merged into a single full-screen rendering pass to reduce intermediate texture read/write and full-screen copy overhead. This work focuses on the rendering algorithm level. To control variables, the experiments use offline eye-tracking replay and simulated ambient light, focusing on the algorithm’s response to ambient illumination and gaze trajectory inputs. Experiments are conducted on a public eye-tracking dataset across multiple scenes. Under strong light in a complex scene, the method improves frame rates by up to 35.73% and 54.71% over fixed foveated rendering and full-resolution rendering, respectively. Except for complex scenes with gradual illumination and random saccades, the coefficients of variation of frame rates under other dynamic test configurations are below 8%. Meanwhile, the ambient-light washout effect effectively masks the objective image quality loss caused by reduced peripheral resolution, making the subjective scores of the proposed method comparable to those of fixed foveated rendering, with even smaller differences under strong light. Ablation studies show that sub-frustum projection, ambient-light adaptive parameter decision, and single-pass full-screen composition each have independent positive contributions and produce synergistic gains. Performance experiment results indicate that the ambient-light washout effect can serve as an auxiliary perceptual basis for peripheral resolution allocation; at the algorithmic design level, it can provide theoretical support for rendering optimization in resource-constrained optical see-through augmented reality.