| 小原 和馬 | 1351028: M, 2回目発表 | 小笠原 司, | 池田 和司, | 高松 淳, | 吉川 雅博 |
|
title: EMG-based hand motion recognition for unspecified users using Deep Neural Networks
abstract: Electromyogram (EMG) is electric potential that occurs during muscle contraction. Wearable human interfaces have been developed using EMG-based hand motion recognition methods. In these methods, there is a problem that recognition rate is low for unspecified users As the cause of the problem, we can mention the effect of the difference between users . For example, differences in muscle mass, differences in body size, and so on. In this presentation, I discuss the method in EMG-based hand motion recognition for unspecified users, which eliminate the effect of the difference between users using Deep Neural Networks. language of the presentation: Japanese | |||||
| 植松 裕介 | 1351011: M, 2回目発表 | 萩田 紀博, | 池田 和司, | 浮田 宗伯 | |
|
title: Human action recognition in still images using weak and semi supervised learning
abstract: In areas such as security, human action recognition in still images is expected for wide applications.Training images are needed to create a model for estimating a human pose. Among various methods for action recognition, those using a human body pose is effective. While the human pose is a crucial cue for action recognition, giving human pose annotations manually to all training images is a huge-time-consuming task. To solve this problem, we propose how to efficiently learn human pose and actions models by integrating and weak and semi supervised learning frameworks. language of the presentation: Japanese | |||||
| 永田 大地 | 1351077: M, 2回目発表 | 安本 慶一, | 池田 和司, | 荒川 豊, | 玉井 森彦 |
|
title: Nap Support System by Smartphone-based Sleep Onset Time Estimation
abstract: In daily life, lack of sleep is one of the main reason for poor concentration. Therefore, A nap is considered as a method to prevent a decrease in work efficiency. For the effective nap, it is necessary to have an appropriate sleep duration. However, if we wake up by using the alarm, wake up time is independent of the actual sleep time from the time fixed. If a person has a longer sleep, he will be more sleepy. In this presentation, we propose a method to estimate a sleep onset time for calculating an appropriate wake up time. Our method only use a smartphone for simplifying the system. We observe a heart rate obtained through a smartphone's camera, and clarify the relationship between a heart rate and a sleep onset time by using a machine learning technique. We report experimental results of sleep state determination for sleep onset time estimation in a nap. We have evaluated our system on experiment that involves 3 subjects and total 8 nap. Our result shows precision ratio of 87% and a recall of 98% for sleep state classification. language of the presentation: Japanese | |||||
| 高木 優 | 1351062: M, 2回目発表 | 松本 裕治☆３, | 池田 和司, | 川人 光男 | |
|
title: Neural Link between Obsessive-Compulsive Disorder and Delay Discounting
We hypothesized that obsessive-compulsive disorder(OCD) is caused by impulsive choices in intertemporal decision problems, e.g.,maladaptive preference of immediate relief from anxiety by compulsions over a future normal life by temporarily restraining their compulsions. We determined whether we could distinguish patients with OCD from healthy controls using functional network related to delay discounting that were obtained by resting-state functional magnetic resonance imaging. First, we detected the functional networks with which we predicted an individual’s discount factor. Then using the detected network, we successfully made a high performance classifier to distinguish patients with OCD from healthy controls. Our findings suggest that OCD might exhibit altered neural organization related to delay discounting language of the presentation: Japanese | |||||
| 福井 辰哉 | 1351092: M, 2回目発表 | 加藤 博一, | 横矢 直和, | Christian Sandor, | 山本 豪志朗 | |
|
title: A Gesture Interface Design for Multiple Users
abstract: With the increase of display size and the improvement of resolution of projectors, we became to use a huge display embironment for showing and controlling digital contents in several users. For controlling digital contents in this embironment, mice-keybords or multi-touch are ofeten used as a input method. However they have problems such as that we need to change a layout of the mice and keyboards or we can't control remote displays. A gesture interface which use RGB-D sensor such as Microsoft Kinect is gathering attention as a new method which solve these problems. Therefore we propose a gesture interface which several users can control digital contents from anywhere in the embironment by solving the current problems of the gesture interface, a occlusion by the other users and a relative cursor indication. language of the presentation: Japanese 発表題目: マルチユーザの為のハンドジェスチャインタフェース設計 発表概要: 液晶ディスプレイの大型化やプロジェクタの解像度向上により，広大なディスプレイを用いて複数のユーザが同時にGUIウィンドウ等のコンテンツを表示，操作するという用途が考えられるようになった．このような環境でコンテンツを操作するために，マルチマウスやマルチタッチからの入力方式がよく用いられているが，それぞれユーザの配置に合わせてキーボード，マウスの配置を変える手間がかかる，操作可能な領域が限られている，といった問題がある．これらの問題を解決するインタフェースとしてMicrosoftが提供するKinectなどのRGB-Dセンサを用いたハンドジェスチャが注目されているが，広大なディスプレイ環境，複数ユーザを仮定した場合には，他ユーザによるオクルージョン，相対座標でのカーソル表示などの問題が生じる．本研究ではこれらの問題を解決するために天井に設置したRGB-Dセンサから取得した深度画像を用いてハンドジェスチャ認識を行い，複数のユーザが任意の位置からコンテンツを操作できるジェスチャインタフェース設計を行う． | ||||||
| 島田 健史 | 1351055: M, 2回目発表 | 小笠原 司, | 横矢 直和, | 高松 淳, | 池田 篤俊 | |
|
title: Cooking Recording System Based on First Person Vision abstract: There is an attempt to digitize of the cooking work in recent studies for education and home robot. First person vision which can obtain gaze information and video information by mounting the camera on the head, because of people operate based on the Image seen with my own eyes when working. Therefore, in this study, we construct the cooking work recording system utilizing First Person Vision. In this presentation, we report experimental results about head of cooks position and orientation estimation, motion detection of near the hands and three-dimensional shape generation of cooking environment utilizing point cloud obtained from RGB-D camera and conditions under which the head and hands are present within a fixed range. language of the presentation: Japanese | ||||||
| 鮫島 一平 | 1361005: D, 中間発表 | 小笠原 司☆４, | 横矢 直和, | 加賀美 聡, | 高松 淳 | |
|
title: Building Environmental Maps of Human Activity for a Mobile Service Robot
abstract: Robotic mapping such as two- or three dimensional occupancy grid map has been a field of study. However such occupancy maps are not enough for advanced tasks in our daily life environments.Therefore, I focus on the following three types of information,1)Human body property,2)human trajectories in 2D that represent how humans move in the environment,3) 3D environmental changes that represents human activities. In this study,I propose a environment maps that are comprised these information.These maps are utilized in order to plan safer, quicker and/or non-human-disturbing paths for a mobile service robot. language of the presentation: Japanese | ||||||
| 松元 裕哉 | 1351099: M, 2回目発表 | 横矢 直和, | 小笠原 司, | 佐藤 智和, | 河合 紀彦, | 中島 悠太 |
|
title: A driving simulation system using novel view synthesis
abstract: Recently, driving-simulation systems for training drivers and developing new vehicles have been developed. Although, the systems imitate the appearance of real traffic environments using 3D models, the created images are different from real scenes because of the artificial texture. The difference makes it difficult for the system to give the feeling of real driving to users. Therefore using photographed images is desirable in driving-simulation systems. However, because a viewpoint moves according to the vehicle movement, which images corresponding to the viewpoints are required. To solve the problem, novel view synthesis, which creates an image different from photographed position, has been proposed. A previous study creates high quality images using 3D point cloud of a real scene and photographed images. However, the computational cost for reconstructing 3D geometry is high. It is difficult for the method to be applied to driving-simulation system because it is necessary to present a novel view image to a user in real-time. This study proposes a method to create a novel view image in real-time by reducing the computational cost, and implement a driving-simulation systems to investigate its validity. language of the presentation: Japanese 発表題目: 自由視点画像生成を用いたドライブシミュレータシステム 発表概要: 現在、運転者の教育や車両開発支援を目的としたドライブシミュレータシステムが開発されている。これらのシステムでは3次元モデルを用いて道路環境を再現しているが、モデルに人工的なテクスチャが使用されるため、実環境とは見え方が異なり、実際の運転で得られる臨場感をユーザに与えることを難しくしている。このため、シミュレータに実写画像を利用することが望まれている。しかし、ドライブシミュレータでは車両の移動に伴う視点の移動が発生し、その位置に対応した視点の画像が必要となる。この問題に対して、撮影地点以外の視点画像を生成する自由視点画像生成手法が研究されている。従来手法では実写画像及びシーンの3次元点群を利用することで高品質な画像を生成しているが、3次元形状の復元に要する計算コストが高いため、リアルタイムに画像を提示する必要のあるドライブシミュレータの用途には不適である。本研究では計算コストを抑制することで自由視点画像をリアルタイムで生成する手法を提案し、実際にドライブシミュレータシステムを実装して画像生成の有効性を調査する。 | ||||||