DehazeGlasses: Optical Dehazing with an Occlusion Capable See-Through Display

DehazeGlasses: Optical Dehazing with an Occlusion Capable See-Through Display
复制标题

DOI:
10.1145/3384657.3384781
复制
发表时间:
2020-03
期刊:
Proceedings of the Augmented Humans International Conference
影响因子:
--
通讯作者:
Yuichi Hiroi;Takumi Kaminokado;Atsushi Mori;Yuta Itoh
Yuichi Hiroi;Takumi Kaminokado;Atsushi Mori;Yuta Itoh
中科院分区:
其他
文献类型:
--
作者:
Yuichi Hiroi;Takumi Kaminokado;Atsushi Mori;Yuta Itoh

文献摘要

相似文献

我们推出 DehazeGlasses,这是一种透明视觉除雾系统,可通过光学方式对用户的视野进行除雾。由于场景环境的各个方面(例如雾霾),人类视力会受到影响。这种退化可能会干扰我们在日常任务中的行为或判断。我们将雾霾场景作为一种常见的退化源,它会由于某些大气条件而使视图变白。与处理记录图像的典型计算机视觉系统不同,我们的目标是实现一种透视眼镜系统,可以光学地操纵我们的视野,以消除感知场景的雾气。我们的系统通过具有遮挡功能的光学透视头戴式显示器 (OST-HMD) 选择性地调节进入眼睛的光强度。我们构建了一个概念验证系统,通过结合数字微镜器件 (DMD) 和 OST-HMD 来评估我们的除雾方法的可行性,并使用用户视角视点相机对其进行测试。对除雾数据集中的 80 个场景进行的定量评估表明,我们的系统实现的去雾视图与感知图像相似性度量下的本机视图相比,明显更接近地面真实场景。此次评估表明,我们的系统实现了感知上自然的去雾效果,同时保持了实际场景的透视效果。
We present DehazeGlasses, a see-through visual haze removal system that optically dehazes the user's field of vision. Human vision suffers from a degraded view due to aspects of the scene environment, such as haze. Such degradation may interfere with our behavior or judgement in daily tasks. We focus on hazy scenes as one common degradation source, which whitens the view due to certain atmospheric conditions. Unlike typical computer vision systems that process recorded images, we aim to realize a see-through glasses system that can optically manipulate our field of view to dehaze the perceived scene. Our system selectively modulates the intensity of the light entering the eyes via occlusion-capable optical see-through head-mounted displays (OST-HMD). We built a proof-of-concept system to evaluate the feasibility of our haze removal method by combining a digital micromirror device (DMD) and an OST-HMD, and tested it with a user-perspective viewpoint camera. A quantitative evaluation with 80 scenes from a haze removal dataset shows that our system realizes a dehazed view that is significantly closer to the ground truth scene compared to the native view under a perceptual image similarity metric. This evaluation shows that our system achieves perceptually natural haze removal while maintaining the see-through view of actual scenes.