360度视频中面向视觉诱发晕动症缓解的实时视口感知光流估计
Real-Time Viewport-Aware Optical Flow Estimation in 360-degree Videos for Visually-Induced Motion Sickness Mitigation
AI总结：
针对VR中视觉诱发晕动症问题，提出基于用户视口的360度视频实时光流估计方法，用于调节颗粒化静止帧不透明度，试点实验显示可能提升观看舒适度。
摘要：
视觉诱发的晕动症（VIMS）是由视觉刺激引起的感知运动副作用，是虚拟现实（VR）广泛使用的主要障碍。与场景物体信息一起，视觉刺激主要通过光流来表征，光流描述了运动模式，如移动图像的强度和方向。我们针对沉浸式用户交互可视化，基于用户当前视口估计360度视频中的实时光流。所提出的方法允许为每个动态360度视频体验估计定制化的视觉流，相比以往对整个等距柱状投影帧考虑单一光流值的方法有所改进。我们将该方法应用于调节颗粒化静止帧（GRFs）的不透明度，该技术由视觉噪声样随机分布的视觉参考组成，在沉浸式预录360度视频体验期间对用户身体保持稳定。我们报告了一项包含18名参与者的试点单次受试者间研究结果，其中用户观看了2分钟的高强度360度视频。结果表明，我们提出的方法成功估计了光流，试点数据显示，GRFs结合实时光流估计可能改善用户观看360度视频时的舒适度。然而，需要更多数据才能获得统计学显著的结果。
英文摘要：
Visually-induced motion sickness (VIMS), a side effect of perceived motion caused by visual stimulation, is a major obstacle to the widespread use of Virtual Reality (VR). Along with scene object information, visual stimulation can be primarily indicated by optical flow, which characterizes the motion pattern, such as the intensity and direction of the moving image. We estimated the real time optical flow in 360-degree videos targeted at immersive user interactive visualization based on the user's current viewport. The proposed method allows the estimation of customized visual flow for each experience of dynamic 360-degree videos and is an improvement over previous methods that consider a single optical flow value for the entire equirectangular frame. We applied our method to modulate the opacity of granulated rest frames (GRFs), a technique consisting of visual noise-like randomly distributed visual references that are stable to the user's body during immersive pre-recorded 360-degree video experience. We report the results of a pilot one-session between-subject study with 18 participants, where users watched a 2-minute high-intensity 360-degree video. The results show that our proposed method successfully estimates optical flow, with pilot data showing that GRFs combined with real-time optical flow estimation may improve user comfort when watching 360-degree videos. However, more data are needed for statistically significant results.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
