| ZHANG YUJIA | M, 2回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 向川 康博, 酒田 信親, 内山 英昭, 磯山 直也 |
|
title:Effect of Relighting in Inpainting
abstract: In contrast to augmented reality, which superimposes CG objects on real space, there is a technology called "Diminished Reality (DR)" that visually erases real objects that exist in reality. In conventional DR, a continuous texture is generated using image inpainting after manually specifying the area to be erased. However, since the three-dimensional spatial structure is not taken into account, the optical integrity of shadows and reflections is not maintained. In this study, we propose a 3D inpainting method that utilizes a machine learning framework to estimate spatial depth from monocular images, thereby realizing DR with optical consistency. language of the presentation:Japanese | |||
| SCHIRM JOHANNES | D, 中間発表 | サイバネティクス・リアリティ工学 | 清川 清, 安本 慶一, 酒田 信親, 内山 英昭, 磯山 直也 |
|
title: Standardizing Quantitative Measures of Presence in Virtual Reality
abstract: "Presence" is a general term for the illusion that a virtual environment is real and the events in it are actually happening. It has been subject to intensive research as long as virtual reality exists. Nowadays, it is commonly evaluated using qualitative measures, which most of the time take the form of questionnaires. However, there have been doubts about whether measures of this type are able to truly capture the concept behind presence. A number of studies propose exploratory techniques based on physiological or behavioral data, but in practice, it is quite difficult to purposefully make use of this type of data in new experimental designs. The proposed research aims to contribute ways of making this easier for as many scenarios as possible. This could be accomplished by first analyzing continuous sensory data like eye-tracking and electromyography using techniques from signal processing, and then building an open-source framework for easy integration into user studies. language of the presentation: English 発表題目: VRにおけるプレゼンスの汎用的定量評価手法の開発 発表概要: 「プレゼンス」とは、ユーザが仮想空間を実在していると認識し、その中に出来事が実際に起こっているように錯覚することの総称である。 しかし、プレゼンスの測定に関する研究はまだまだ途上段階である。 現在では、アンケート形式による定性的な尺度での評価が一般的である。 しかし、プレゼンスの測定に定性的尺度が最適であるかどうかについては、まだ議論が行われている段階である。 行動や生理などに関する定量的データに基づく手法は数多くあるが、それらのデータを様々な実験において汎用的に利用するツールは、まだ開発されていない。 そこで、本研究では、幅広いシナリオでの定量的データを容易に活用できる方法の提案を目的とする。 具体的には、アイトラッキングや筋電図などのデータを信号処理の技術を用いて解析し、実験に簡単に組み込むことができるオープンソースのフレームワークを構築する。 | |||
| LIU HUAKUN | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 安本 慶一, 内山 英昭, 磯山 直也 |
|
title: Adaptive Data-driven Inertial Pedestrian Dead Reckoning using Head-mounted IMU
abstract: The inertial pedestrian dead reckoning (IPDR) has been widely used to provide self-contained and continuous position of pedestrian for location based services, which is one of the key of ubiquitous system. But the strict constrain or low-accuracy has greatly limited the usage of such system. Recently, there has been a growing interest in applying deep neural networks to breakthrough current bottleneck, which is called data-driven IPDR. However, the performance of current low generalizable data-driven IPDR has been limited by the lack of diversity motion and device datasets. In this work, we try to design an adaptive data-driven inertial IPDR system to improve the generalizability, make the trained system easily transferred and applied for any device, any pedestrian motion. Moreover, we design it using the specific head-mounted IMU, which can facilitate the development of smart glasses-based location services. language of the presentation: English | |||
| LU JINZHENG | M, 1回目発表 | 光メディアインタフェース | 向川 康博, 中村 哲, 舩冨 卓哉, 品川 政太朗, 藤村 友貴 |
|
title: Automatic colorization for anime trace image abstract: Automatic colorization is a significant task especially for Anime industry. The purpose of this study is to realize an automatic Anime colorization method, in which multiple images can be colored with reference images. The existing work colorize line drawing one by one, the improvement of animation production efficiency is limited. We try to use the image feature matching technique to make the reference image select the object of temporal consistency in successive frames, and realize automatic colorization. language of the presentation: English | |||