| 黒川 陽平 | 1351040: M, 2回目発表 | 横矢 直和, | 加藤 博一, | 佐藤 智和, | 中島 悠太 |
|
title: Camera pose estimation without similarity-based feature point matching
abstract: Precise camera pose estimation is required for mobile AR in a known scene, which presents information based on a location of the mobile devices. Recently, some methods have been proposed using a preliminarily built 3D point cloud database. They estimate camera poses by solving the PnP problem with a set of 2D-3D matches between feature points in the query image and 3D points in the database. The 2D-3D matches can be obtained by using a similarity between descriptors for a 2D point and a 3D point. However, these methods does not work on mobile devices due to their high computational cost caused by descriptor extraction/matching. Also, they require a large database storing millions of 3D points each of which is associated with a descriptor. The goal of this research is to propose a light-weight camera pose estimation method that does not need to make explicit 2D-3D matches based on descriptors. In this talk, we propose a new error criterion for given camera poses without any similarity-based matching, and show an experimental result for validating the behavior of the criterion. language of the presentation: Japanese | |||||
| 黄 頌友 | 1351124: M, 2回目発表 | 横矢 直和, | 加藤 博一, | 佐藤 智和, | 中島 悠太 |
|
title: 3D Reconstruction of a Handheld Object with a Fixed RGB-D Camera abstract: Recently, some methods have been proposed that reconstruct a 3D geometry model of a handheld rigid object, where the object is grasped and moved so that it can be captured from all directions with a fixed RGB-D camera. To reconstruct a 3D geometry model, registration of object point clouds obtained from different RGB-D frames is required. However, registration in these methods require gloves or special hardware to distinguish the object from the hands. In this work, considering that the object pose can be described by rigid motion while the hands cannot, we propose to use feature points extracted from RGB-D frames and apply RANSAC to obtain initial estimates of the object pose. We use ICP to refine the initial pose estimates. Then, after removing the outlier points, which can be considered as ones on the hand, we are able to build a entire 3D geometry model of the object. In this presetation, we show our progress in the result of initial estimates. language of the presentation: Japanese | |||||
| 吉本 公則 | 1351111: M, 2回目発表 | 小笠原 司, | 加藤 博一, | 高松 淳, | 吉川 雅博 |
|
title: Classifying types of grasp using topological index
abstract: In this research, we propose a method for classifying the types of grasp. Such estimation is employed to map the grasp obtained from human demonstration into the grasp for a robot hand that has different shape. Since the type of grasp is defined from not metric, rather topological aspects, we propose to use the Gauss Linking Integral as descriptor of grasp. We use the Support Vector Machine to classify fourteen types of grasp, which are originated from taxonomy of grasp in everyday life proposed by Kamakura et al. We actually evaluate the accuracy of the proposed method. language of the presentation: Japanese | |||||
| JAKOVLJEVIC NEMANJA | 1351120: M, 2回目発表 | 小笠原 司, | 加藤 博一, | 高松 淳, | 吉川 雅博 |
|
title: Per robot path finding from UAV measurements
abstract: Various robots are starting to be a part of our daily life. In order for them to move efficiently in our environment, we need to have the knowledge about surface on which they will be moving. In this research, I am developing a system that would provide a way to rapidly model the environment and for finding the path between given points suitable for a selected robot. language of the presentation: English | |||||
| PADERNA RYAN R. | 1351123: M, 2回目発表 | 岡田 実, | 杉本 謙二, | 東野 武史 | ||
|
title: Modified Orthogonal Matching Pursuit based Channel Estimation for ISDB-T abstract:This research proposed a channel estimation for ISDB-T using Modified Orthogonal Matching Pursuit. The proposed method is capable is estimating the channel in doubled selective fading channel and fractional multipath channel. This system uses a signal oversampling that would give robustness against fractional delay. In addition, the proposed system maintains the computation time in estimating the channel even though oversampling was implemented. language of the presentation: English | ||||||
| REINOSO CHISAGUANO DIEGO JAVIER | 1361022: D, 中間発表 | 岡田 実, | 杉本 謙二, | 東野 武史 | ||
|
title: MIMO-OFDM and ISDB-T receivers with ESPAR antenna-based diversity
abstract: Diversity is one of the techniques used in wireless communication systems to combat the effects of multipath fading and improve the performance. We present the design of MIMO-OFDM and ISDB-T receivers that use an ESPAR antenna to achieve diversity gain without increasing the hardware complexity. Two channel estimation methods are presented for the MIMO-OFDM receiver. We show that the compressed sensing-based channel estimation achieves a high accuracy with low computational cost. The simulation results show that our proposed receivers are able to achieve a substantial improvement in the bit error rate performance. language of the presentation: English | ||||||
| 大島 悠司 | 1351016: M, 2回目発表 | 中村 哲, | 杉本 謙二, | 戸田 智基, | Sakriani Sakti, | Graham Neubig |
|
title: Prosody Correction Preserving Speaker Individuality in English-Read-By-Japanese Speech Synthesis
abstract: To synthesis English speech preserving Japanese speaker individuality, it is effective to use an English-Read-by-Japanese. However, synthetic speech is not enough naturalness as an English speech because it is reflected an English-Read-by-Japanese. In this research, we propose a method to correct prosody of English-Read-by-Japanese using model adaptation method based on the HMM speech synthesis. The experimental result show that we can synthesis English speech that improve naturalness preserving Japanese speaker individuality by the proposed method. language of the presentation: Japanese 発表題目: 日本人英語音声合成における話者性を保持した韻律補正 発表概要: 日本語母語話者の話者性を保持した英語音声合成を実現する上で、日本人英語の利用が有効である。しかしながら、合成音声は日本人英語の特徴を捉えたものとなるため、英語音声として十分な自然性が得られないという問題がある。これに対し本稿では、HMM音声合成におけるモデル適応法を用いて、英語母語話者の韻律情報に基づき韻律を補正する手法を提案する。実験結果から、提案法により話者性を保持しつつ、自然性を改善できることを示す。 | ||||||
| 西垣 友理 | 1351081: M, 2回目発表 | 中村 哲, | 杉本 謙二, | 戸田 智基, | Sakriani Sakti, | Graham Neubig |
|
title: HMM-Based Speech Synthesis Controling with Speech and Text
abstract: Development of Corpus-based Speech Synthesis, Speech Synthesis which has specific character voice was built and we looking foward to using that technology at derived work using Speech synthesis. Specially, HMM-Based Speech Synthesis that can synthesize speech from text can control synthetic speech parameter easier. In this talk, we will proposed HMM-Based Speech Synthesis controling with speech and text. We investigate and evaluate the voice unit which is used in imitation of the input meter. We propose more higher nature-related and correspondence to independent speaker. language of the presentation: Japanese 発表題目: HMM音声合成における音声・テキストを用いた制御法 発表概要: コーパスベース音声合成技術の発達により，特定のキャラクタ性を有する音声合成技術が構築され，所望の音声を創作する活動においてその利用が期待されている．特にテキストから音声を合成する技術の一つであるHMM音声合成は，合成音声の特徴を柔軟に制御することが可能である．本発表では，HMM音声合成において，通常のテキスト音声合成の機能を保持し，かつ，音声を用いて合成音声の韻律を制御する手法を提案する．本研究では、入力されたF0系列を模倣する際の音声単位の検討および評価を行った. このことより，合成音声の更なる自然性の改善と任意の話者への対応を目指す． | ||||||
| 上田 健揮 | 1351010: M, 2回目発表 | 安本 慶一, | 藤川 和利, | 荒川 豊, | 玉井 森彦 |
|
title: A System for Daily Living Activities Recognition Based on Multiple heterogeneous Sensing Data in a Smart Home
abstract: In this study, we propose a daily living activity recognition method with low initial costs and low privacy exposure by using only power meters and indoor positioning sensors that are expected to widespread in the future. In the proposed method, we extract training data for activities from sensor data measured in a smart house testbed, by using a labeling tool we developed, and construct a model for classifying activities by machine learning. As evaluation, we constructed a model targeting 5 activities using SVM and confirmed that the model can classify activities with about 85.05% precision on average. language of the presentation: Japanese 発表題目: スマートホームにおける複数異種センシングデータに基づいた生活行動データ抽出システムの提案 発表概要: 本研究では，今後普及することが見込まれる消費電力センサと位置情報センサのみを 用いることで，導入コストが低くプライバシーの侵害が少ない生活行動認識手法を提案する． 提案手法では，開発した生活行動ラベリングツールを用いて訓練データを取得し， 機械学習により行動認識モデルを構築する．評価実験として，５種類の行動を対象としてSVM によりモデルを構築し，平均85.05%の適合率を得ることができた． | |||||
| 杉田 敢 | 1351060: M, 2回目発表 | 安本 慶一, | 藤川 和利, | 荒川 豊, | 玉井 森彦 |
|
title: A method for non-invasively estimating hunger degree and blood glucose based on meal and exercise logs
abstract: If temporal variation of a person’s hunger degree could be estimated, it would be possible to adjust his/her eating habits and/or prevent obesity. It is well-known that there is a negative correlation between a hunger degree and a blood glucose level. However, it is hard to measure a person’s blood glucose level anytime and anywhere, because it relies usually on an invasive method (e.g., blood sampling). This study proposes a method for non-invasively estimating a person's hunger degree and blood glucose based on meal and activity logs, and investigates availability and estimation accuracy of this method. language of the presentation: Japanese 発表題目: 食事・行動履歴に基づく非侵襲的血糖値・空腹度推定手法の提案 発表概要: 人間の空腹レべルの時間的変化が推定できれば，食習慣の調整や肥満の防止に役立てる事が可能となる．空腹レベルと血糖値との間には負の相関が存在することが一般的に知られている．しかし，人間の血糖値の測定は採血を伴う侵襲的な測定に依るため，任意の時刻・場所において測定することは困難である．本研究では，食事・行動履歴を基に，人間の血糖値および空腹レベル(空腹度)を非侵襲的に推定する手法を提案し，手法の推定精度と有用性を調査する． | |||||
| 阪口 紘生 | 1351047: M, 2回目発表 | 伊藤 実, | 藤川 和利, | 柴田 直樹 | |
|
title: A Reliable Routing in Vehicular Networks using Roadside Boxes abstract: In VANET, Using Geometric Routing scheme is a mainstream. However, it is difficult to deliver a packet to the destination under the low density environment. In low density environment, Some Delay Tolerant method are suggested, but is not practical without being able to guarantee the packet arrival rate. In this work, we introduce wireless base stations without backbone on the roadside. And we consider a car to be a carrier and realize the multi-hop communication between a base station and the base station. This is a new network form. We propose a routing method in like this network. Past study, we propose a method to minimize car number to use by a destination and to consider traffic to be a band, and to choose the best band.These method is established only by a car and the communication of the base station. Therefore we suggest the method that the communication between the roadside boxes uses VANET method. From this We establish method that cleared problems in low density area and high density area. Recently, I perform an experiment to compare the precedent technique with the suggestion method to check the effectiveness of the suggestion method. Future study, we should compare suggestion method with other VANET Routing method and Epidemic Routing. language of the presentation:Japanese | |||||
| 鷲尾 直大 | 1351113: M, 2回目発表 | 藤川 和利, | 伊藤 実, | 安本 慶一, | 猪俣 敦夫 |
|
title: A Proposal of the Vehicle Clustering Method for Information Propagation abstract: In order to realize the driving route recommendation and autonomous driving for the burden of the driver, information sharing among vehicles is essential. In addition, it is believed that in order to share information in real time about road congestion and traffic accidents occur suddenly, the information propagation using inter-vehicle communication is effective. However, traditional routing methods and vehicle clustering methods for the purpose of information sharing do not consider vehicle trajectories in passing communication. As a result, it causes a decrease in the efficiency of information propagation. For the purpose of information propagation efficiently, we propose a vehicle clustering method based on the similarity of geographic position and vehicle trajectories. language of the presentation: Japanese 発表題目: 車車間通信における情報伝播のためのクラスタリング手法の提案 発表概要: 運転者の負担軽減のための自動運転や走行経路推薦を実現するためには，車車間での情報共有が必要不可欠である．そして，突発的に発生する事故や渋滞をリアルタイムに情報共有するためには，車車間通信を利用した情報伝播が効果的であると考えられる．しかしながら，情報共有を目的とした従来のルーティング及び車両クラスタリング手法では，対向車両とのすれ違い通信時に走行軌跡を考慮しておらず，結果として情報の伝播効率の低下を引き起こす．そこで本研究では，効率的な情報伝播を目指し，地理位置と走行軌跡の類似性に基づく車両クラスタリング手法を提案する． | |||||