¥¼¥ß¥Ê¡¼¥ëȯɽ

Æü»þ: 9·î27Æü(²Ð)1¸Â¡Ê9:20-10:50¡Ë


²ñ¾ì: L£±

»Ê²ñ: µ×ÊÝ ¿ÒÇ·
¼¯Åç ¾Ä 1551032: M, 2²óÌÜȯɽ ¥¤¥ó¥¿¥é¥¯¥Æ¥£¥Ö¥á¥Ç¥£¥¢Àß·×³Ø ²ÃÆ£ Çî°ì,¸þÀî ¹¯Çî,Christian Sandor,ÉðÉÙ µ®»Ë
title: Image Generation Corresponding to the Eye Movement for Head-Mounted Displays Based on Microlens Arrays and it's Evaluation
abstract: Recently, various head-mounted displays (HMDs) are being popularized. HMDs which is thin and with characteristics of the wide field of veiw are still not practical. We can make them with the method of using microlens arrays. However, we should generate display images corresponding to the eye movement when we use that HMDs with wide field of view. In this research, I suggest the system of image generation corresponding to the eye movement, and I evaluate it.
language of the presentation: Japanese
ȯɽÂêÌÜ: ¥Þ¥¤¥¯¥í¥ì¥ó¥º¥¢¥ì¥¤¤òÍѤ¤¤¿HMD¤Ë¤ª¤±¤ë´ãµå°ÜÆ°¤ËÂбþ¤·¤¿²èÁüÀ¸À®¤È¤½¤Îɾ²Á
ȯɽ³µÍ×: ¶áǯ¡¢Â¿Íͤʥإåɥޥ¦¥ó¥È¥Ç¥£¥¹¥×¥ì¥¤¡ÊHMD)¤¬¿Í¡¹¤Ë¹­¤Þ¤ê¤Ä¤Ä¤¢¤ë¡£¤Þ¤¿¡¢Çö¤¯¤Æ»ëÌî³Ñ¤ÎÂ礭¤¤HMD¤Ï¡¢Ì¤¤À¼ÂÍÑŪ¤Ë¤Ê¤Ã¤Æ¤¤¤Ê¤¤¡£Çö¤¯¤Æ»ëÌî³Ñ¤ÎÂ礭¤¤HMD¤Ï¡¢¥Þ¥¤¥¯¥í¥ì¥ó¥º¥¢¥ì¥¤Êý¼°¤òÍѤ¤¤ë¤³¤È¤Ë¤è¤ê¡¢¼Â¸½¤¹¤ë¤³¤È¤¬¤Ç¤­¤ë¡£¤·¤«¤·¤Ê¤¬¤é¡¢¹­»ëÌî³Ñ¤Ê¥Þ¥¤¥¯¥í¥ì¥ó¥º¥¢¥ì¥¤Êý¼°¤òÍѤ¤¤¿HMD¤Ë¤ª¤¤¤Æ¡¢¥Ç¥£¥¹¥×¥ì¥¤¾å¤Î²èÁüÀ¸À®¤ò´ãµå°ÜÆ°¤ËÂбþ¤µ¤»¤ëɬÍפ¬¤¢¤ë¡£¤½¤³¤ÇËܸ¦µæ¤Ç¤Ï¡¢´ãµå°ÜÆ°¤ËÂбþ¤·¤¿²èÁüÀ¸À®¤Î¥·¥¹¥Æ¥à¤ÎÄó°Æ¤È¡¢¤½¤Îɾ²Á¤ò¹Ô¤¦¡£
 
»³ËÜ Íºµ® 1551115: M, 2²óÌÜȯɽ ¥¤¥ó¥¿¥é¥¯¥Æ¥£¥Ö¥á¥Ç¥£¥¢Àß·×³Ø ²ÃÆ£ Çî°ì,¸þÀî ¹¯Çî,Christian Sandor,ÉðÉÙ µ®»Ë

title: *** Study on Object-based Visual SLAM ***

abstract: *** Visual SLAM is a technique for estimating the position and orientation of a camera in 3D by analyzing the sequentially taken pictures from a moving camera.

In general, data becomes enormous as a result of continous SLAM usage.

And it replaces objects to 3D models prepared in advance in order to reduce datas.

Furthermore, uses the positional relationship of the object to eliminate the error.

However, it requires to create enormous 3D models in advance in order to apply to a variety of environments.

The biggest strength of SLAM is that can be apply to unknown environments, but SLAM++ does not have it.

Therefore, The purpose of this study is to investigate a SLAM algorithm based on the object does not need to prepare a 3D models in advance.

As the pre-stage of the study, we replaced object recognition part of SLAM++ algorithm for recognition of artificial markers, and evaluated how it can reduce the errors.

In this presentation, I will report about the results.***


language of the presentation: *** Japanese ***

ȯɽÂêÌÜ: *** ¥ª¥Ö¥¸¥§¥¯¥È¤Ë´ð¤Å¤¯SLAM¥¢¥ë¥´¥ê¥º¥à¤Î¸¡Æ¤***

ȯɽ³µÍ×: *** Visual SLAM(Simultaneous Localization and Mapping)¤ÏÆ°²èÁü¤ò²òÀϤ¹¤ë¤³¤È¤Ç¡¤¥ê¥¢¥ë¥¿¥¤¥à¤Ë¥«¥á¥é¤Î°ÌÃÖ¡¦»ÑÀª¤È¥·¡¼¥ó¤Î»°¼¡¸µ¾ðÊó¹½Â¤¤ò¿äÄꤹ¤ë¼êË¡¤Ç¤¢¤ë¡¥

°ìÈ̤ËSLAM¤Ë¤è¤ë¿äÄê¤ò³¤±¤ë¤È,»°¼¡¸µ¹½Â¤¤òÊÝ»ý¤¹¤ë¤¿¤á¤Î¥Ç¡¼¥¿Î̤¬ËÄÂç¤Ë¤Ê¤ë¤È¸À¤¦ÌäÂêÅÀ¤¬¤¢¤ë.¤³¤Î¤è¤¦¤ÊÌäÂê¤ËÂФ·¤Æ,»°¼¡¸µÊªÂÎǧ¼±¤Îµ»½Ñ¤òSLAM¥¢¥ë¥´¥ê¥º¥à¤Ë¼è¤êÆþ¤ì¤ë¤³¤È¤Ë¤è¤ê,ǧ¼±¤µ¤ì¤¿ÊªÂΤÎÎΰè¤ò»öÁ°¤ËÍÑ°Õ¤·¤¿»°¼¡¸µ¥â¥Ç¥ë¤ÇÃÖ¤­´¹¤¨¤ë¤³¤È¤Ë¤è¤ê¥Ç¡¼¥¿Î̤òºï¸º¤·¤Æ¤¤¤ë.¤Þ¤¿,¥ª¥Ö¥¸¥§¥¯¥È¤È¥«¥á¥é¤Î°ÌÃÖ´Ø·¸¤ò»ÈÍѤ¹¤ë¤³¤È¤Ë¤è¤êÃ༡¿äÄê½èÍý¤Ë¤è¤êÃßÀѤ¹¤ë¸íº¹¤Î²ò¾Ã¤â²Äǽ¤È¤·¤Æ¤¤¤ë.¤·¤«¤·¡¤ÍÍ¡¹¤Ê´Ä¶­¤ËÂбþ²Äǽ¤È¤¹¤ë¤¿¤á¤Ë¤ÏËÄÂç¤Ê»°¼¡¸µ¥â¥Ç¥ë¤ò»öÁ°¤ËÍÑ°Õ¤¹¤ëɬÍפ¬¤¢¤ë¡¥¤³¤ì¤Ï¡¤SLAM¤ÎºÇÂç¤ÎÍøÅÀ¤Ç¤¢¤ë»öÁ°¤Î½àÈ÷¤òɬÍפȤ»¤º¤ËÍÍ¡¹¤Ê´Ä¶­¤ÇÍøÍѲÄǽ¤È¤¤¤¦ÍøÊØÀ­¤ò»¤Ê¤Ã¤Æ¤·¤Þ¤¦¡¥¤½¤³¤Ç¡¤Ëܸ¦µæ¤Ç¤Ï¡¤»öÁ°¤Ë»°¼¡¸µ¥â¥Ç¥ë¤ò½àÈ÷¤¹¤ëɬÍפ¬¤Ê¤¤¥ª¥Ö¥¸¥§¥¯¥È¤Ë´ð¤Å¤¯SLAM¥¢¥ë¥´¥ê¥º¥à¤Î¸¡Æ¤¤ò¹Ô¤¦¡¥ËÜȯɽ¤Ç¤Ï¡¤»öÁ°Ãʳ¬¤È¤·¤Æ¡¤SLAM++¤Î¥¢¥ë¥´¥ê¥º¥à¤ÎʪÂÎǧ¼±Éôʬ¤ò¿Í¹©¥Þ¡¼¥«¤Îǧ¼±¤ËÃÖ¤­´¹¤¨¤¿¾å¤Ç¡¤ÊªÂÎǧ¼±¤Ë¤è¤ëÃßÀÑ¸íº¹¤ÎÄ㸺¸ú²Ì¤Ë¤Ä¤¤¤Æ¸¡¾Ú¤ò¹Ô¤Ã¤¿¤Î¤ÇÊó¹ð¤¹¤ë¡¥ ***

 
º´Æ£ ÂçÌ´ 1551049: M, 2²óÌÜȯɽ ¸÷¥á¥Ç¥£¥¢¥¤¥ó¥¿¥Õ¥§¡¼¥¹ ¸þÀî ¹¯Çî,²ÃÆ£ Çî°ì,çÕÉÚ ÂîºÈ,µ×ÊÝ ¿ÒÇ·
title: Investigation of visual expression for non-planar display
abstract:
language of the presentation: Japanese
ȯɽÂêÌÜ¡§ÈóÊ¿Ì̥ǥ£¥¹¥×¥ì¥¤¤Î¤¿¤á¤Î±ÇÁüɽ¸½¤ÎÄ´ºº
ȯɽ³µÍס§¡Ê´ë¶È¤È¤Î¶¦Æ±¸¦µæ¤Î¤¿¤áÈó¸ø³«¡Ë
 
Éú²¼ ¿¸ 1551090: M, 2²óÌÜȯɽ ¸÷¥á¥Ç¥£¥¢¥¤¥ó¥¿¥Õ¥§¡¼¥¹ ¸þÀî ¹¯Çî,²ÃÆ£ Çî°ì,çÕÉÚ ÂîºÈ,µ×ÊÝ ¿ÒÇ·
title: Reproducing optical properties of translucent objects by mixing scattering liquids
abstract: Replica of translucent objects is highly demanded in various fields such as realistic skin reproduction. In this research, we propose a method to reproduce translucency by mixing several types of liquid. First, we observe different scattering from different mixture ratios of translucent liquids in order to find out the relationship. We verify the linear relationship between mixture ratio and scattering. Then, we attempt to design a system to estimate optimal mixture ratio by the acquired linear relationship in order to reproduce translucent objects.
language of the presentation: Japanese
ȯɽÂêÌÜ: ±ÕÂκ®¹ç¤Ë¤è¤ëȾƩÌÀʪÂΤθ÷³ØÆÃÀ­¤ÎºÆ¸½
ȯɽ³µÍ×: ȾƩÌÀʪÂΤΥì¥×¥ê¥«¤ÏÃé¼Â¤ÊÈ©¤ÎºÆ¸½¤Ê¤É¡¢ÍÍ¡¹¤ÊʬÌî¤Ç¼ûÍפ¬¹â¤Þ¤Ã¤Æ¤¤¤ë¡£Ëܸ¦µæ¤Ç¤ÏÊ£¿ô¤Î±ÕÂΤκ®¹ç¤Ë¤è¤Ã¤Æ¡¢È¾Æ©ÌÀ´¶¤Î¸÷³ØÆÃÀ­¤Ë´ð¤Å¤¤¤¿ºÆ¸½¤òÌܻؤ¹¡£¤Þ¤º¡¢È¾Æ©ÌÀ´¶¤Ë±Æ¶Á¤òÍ¿¤¨¤ë°Û¤Ê¤ëÀ­¼Á¤ò»ý¤Ä±ÕÂΤκ®¹çÈæΨ¤Ë¤è¤ë¥Ð¥ê¥¨¡¼¥·¥ç¥ó¤ò³Îǧ¤·¤¿¡£¤¹¤ë¤È¡¢º®¹çÈæΨ¤ÈȾƩÌÀ´¶¤Î´Ö¤ËÀþ·ÁÀ­¤¬³Îǧ¤Ç¤­¤¿¡£¤½¤³¤Ç¡¢Ç¤°Õ¤ÎȾƩÌÀʪÂΤòºÆ¸½¤¹¤ë¤¿¤á¡¢Àþ·ÁÀ­¤òÍøÍѤ·¤¿ºÇŬ¤Ê±ÕÂΤκ®¹çÈæΨ¤Î¥Ñ¥é¥á¡¼¥¿¤ò¿äÄꤹ¤ë¥·¥¹¥Æ¥à¤òÄó°Æ¤¹¤ë¡£
 

²ñ¾ì: L2

»Ê²ñ: Duong Quang Thang
ÂçÆâ ·¼¼ù 1561004: D, Ãæ´Öȯɽ ¼«Á³¸À¸ì½èÍý³Ø ¾¾ËÜ Íµ¼£,Ãæ¼ ů,¿·ÊÝ ¿Î,¿ÊÆ£ ͵Ƿ,ǽÃÏ ¹¨

title: Higher-Order Modeling for Predicate Argument Structure Analysis

abstract: Existing methods for Japanese predicate argument structure (PAS) analysis identify case arguments of each predicate without considering interactions between the target PAS and others in a sentence. However, the argument structures of the predicates in a sentence are semantically related to each other. In this work, we propose new methods for Japanese PAS analysis to jointly identify case arguments of all predicates in a sentence by (1) modeling multiple PAS interactions and (2) approximately searching optimal PAS combinations. Performing experiments on the NAIST Text Corpus, we show that our joint analysis methods achieve the state-of-the-art results.

language of the presentation: Japanese

 
Âç¼ Éñ 1561006: D, Ãæ´Öȯɽ ¼«Á³¸À¸ì½èÍý³Ø ¾¾ËÜ Íµ¼£,Ãæ¼ ů,¿·ÊÝ ¿Î,¿ÊÆ£ ͵Ƿ,ǽÃÏ ¹¨
title: Information structure analysis of abstracts in multiple domains using word embeddings.
abstract: We present an automated approach to identify information structure with sentence in abstract like background, objective, method and results. Structured Abstract using general medical domain can be utilized training model to identify the information structure. However, it is difficult to adapt the model to another domain (e.g. NLP domain). We show results on methods to identify information structured sentence using word embeddings and the effective of word embeddings on multi-domain.
language of the presentation: Japanese
ȯɽÂêÌÜ: Ê£¿ô¤Î¥É¥á¥¤¥ó¤Ë¤ª¤±¤ë¥¢¥Ö¥¹¥È¤Î¹½Â¤²òÀϤÈWord embeddings¤Î±Æ¶Á
ȯɽ³µÍ×: ËÜȯɽ¤Ç¤Ï¥¢¥Ö¥¹¥È¥é¥¯¥È¤Î¼«Æ°¹½Â¤²½²òÀϼêË¡¤Ë¤Ä¤¤¤Æȯɽ¤¹¤ë¡£°åÎŷϤÎÏÀʸ¤Ç¤Ï ¤½¤Î¸¦µæ¤ÎÇØ·Ê¡¢¼êË¡¡¢·ë²Ì¤Ê¤É¤ò¹½Â¤²½¤·¤Æ½ñ¤«¤«¤ì¤¿¹½Â¤²½¥¢¥Ö¥¹¥È¥é¥¯¥È¤¬ºÎÍѤµ¤ì¤Æ¤ª¤ê¡¢Æɤߤ䤹¤¤ ¹½À®¤È¤Ê¤Ã¤Æ¤¤¤ë¡£¤Þ¤¿¡¢¤³¤Î¥¢¥Ö¥¹¥È¥é¥¯¥È¤Î¹½Â¤¤Ï¤µ¤Þ¤¶¤Þ¤Ê²Ê³Øµ»½Ñ·ÏÏÀʸ¤Ë´Ø¤¹¤ë¼«Æ°Í×Ìó¡¢QA¥¿¥¹¥¯¤È¤¤¤Ã¤¿²¼Î®¥¿¥¹¥¯ ¤Ë¤ª¤¤¤ÆÍ­¸ú¤Ê¾ðÊó¤Ç¤¢¤ë¤È¤¤¤¦¤³¤È¤¬¤ï¤«¤Ã¤Æ¤¤¤ë¡£¹½Â¤²½¥¢¥Ö¥¹¥È¥é¥¯¥È¤ò¶µ»Õ¥Ç¡¼¥¿¤È¤·¤ÆÍѤ¤¤ë¤³¤È¤Çµ¡³£³Ø½¬¤Î¼êË¡¤Ë¤è¤ê ¼«Æ°¹½Â¤²½¥â¥Ç¥ë¤ò¹½ÃÛ¤¹¤ë¤³¤È¤Ï²Äǽ¤Ç¤¢¤ë¡£¤·¤«¤·¹½Â¤²½¥¢¥Ö¥¹¥È¥é¥¯¥È¤Ï°å³Ø·Ï¤ÎʬÌî¤Ç¤ÏºÎÍѤµ¤ì¤Æ¤¤¤ë ¤â¤Î¤Î¡¢Î㤨¤Ð¼«Á³¸À¸ì½èÍý¤ÎʬÌî¤Ê¤É¤Ç¤Ï¤¢¤Þ¤êºÎÍѤµ¤ì¤Æ¤ª¤é¤º¡¢ÍѸì¤Î°ã¤¤¤Ê¤É¤Ç¥â¥Ç¥ë¤ÎÀºÅ٤˱ƶÁ¤¬¤Ç¤ä¤¹¤¤¡£ ¤½¤³¤Ç¥é¥Ù¥ë¤Ê¤·¤Ç¤â¹½ÃۤǤ­¤ëword embedding¤òÍѤ¤¤ë¤³¤È¤Ç¤É¤ÎÄøÅÙ¥â¥Ç¥ë¤ÎÀºÅÙ¸þ¾å¤Ë´óÍ¿¤Ç¤­¤ë¤«¤ò ¸¡Æ¤¤¹¤ë¡£
 

²ñ¾ì: L3

»Ê²ñ: Ȫ ½¨ÌÀ
ÂçÆâ ΤºÚ 1551021: M, 2²óÌÜȯɽ ¿ôÍý¾ðÊó³Ø ÃÓÅÄ Ï»Ê,¶âë ½Åɧ,µ×ÊÝ ¹§ÉÙ
title: Extraction of interaction signals between humans and dogs
abstract: Interaction is an essential skill of human being. Dogs have also acquired their interaction skills with humans through the domestication. However, the signals of interaction between humans and dogs are unclear. To elucidate the signals, we analyzed their behaviors using an accelerometer and an eye tracker. Our result shows that the accelerometer data includes the information of emotional states and that dog specialists extract some information from dogs¡Ç behavior by gazing important points.
language of the presentation: Japanese
 
CHEN RONGXIU 1551122: M, 2²óÌÜȯɽ ¿ôÍý¾ðÊó³Ø ÃÓÅÄ Ï»Ê,¶âë ½Åɧ,º´¡¹ÌÚ Çî¾¼
¥¿¥¤¥È¥ë¡§ ¿À·Ð¶ËÀ­·ÁÀ®¤Î¿ôÍýŪ²òÀÏ
ȯã²áÄø¤Ë¤è¤Ã¤Æ¿À·ÐºÙ˦¤Î·ÁÂÖ¤¬¶ËÀ­¤òÀ¸¤ß½Ð¤¹¡£Âè°ìÃʳ¬¤Ç¤Ï¡¢¥Ë¥å¡¼¥í¥ó¤Ï¡¢¿À·ÐÆ͵¯¤È¸Æ¤Ð¤ì¤ë¤¤¤¯¤Ä¤«¤Î±ß·Á¤Î¤è¤¦¤ÊÉôʬ¤ò»ý¤Ã¤Æ¤¤¤ë¡£¿À·ÐÆ͵¯¤¬»÷¤Æ¤¤¤ë¤Ë¤â¤«¤«¤ï¤é¤º¡¢¼¡¤ÎÃʳ¬¤Ç¡¢¿À·ÐÆ͵¯¤Î°ì¤Ä¤Ï¡¢Â¾¤Î¤è¤êŤ¯¿­¤Ó¤ë¡£¤³¤ÎºÇĹ¤Î¿À·ÐÆ͵¯¤¬¼´º÷¤È¤Ê¤ê¡¢¤Û¤«¤ÎÆ͵¯¤¬¡¢¾­ÍèŪ¤Ë¤Ï¼ù¾õÆ͵¯¤Ë¤Ê¤ë¡£»ä¤¿¤Á¤Ï¡¢¤³¤ÎÂоÎÀ­¤òÇˤì¿À·Ð¥×¥í¥»¥¹¤ò¡Ö¿À·Ð¶ËÀ­·ÁÀ®¡×¤È¸Æ¤ó¤Ç¤¤¤ë¡£¤³¤ÎÂоÎÀ­Ç˲õ¤Î²áÄø¤Ï¡¢¿À·ÐºÙ˦¤¬¼ù¾õÆ͵¯¤È¼´º÷¤È¤¤¤¦Æ͵¯¤Îµ¡Ç½Ê¬Î¥¤ò¤¹¤ë¤¿¤á¤Ë½ÅÍפǤ¢¤ë¡£
Ëܸ¦µæ¤ÎÌÜŪ¤ÏÆ͵¯¿­Ä¹°Ý»ý¤¹¤ë¸¶Íý¤ÈÈóÂоξõÂ֤ΰÂÄêÀ­¤¹¤ë¿ôÍý¤ò²òÌÀ¤¹¤ë¤³¤È¤Ç¤¹¡£
Title:Mathematical interpretation of neural polarization
Abstract:Neurons in a developing stage changes morphology dramatically. In the first phase, a neuron has circular shape or several protrusions called as neurite. Neurites are similar¡¡in length in this phase, nevertheless, one of the neurites elongates longer than others in the following phase. This longest neurite becomes axon and the others become dendrite in future. We call this symmetry breaking process "neural polarization", which is important for neurons to decide neurite functions.
There are two goals for this work. One is to interpret neuritis growing, while another is to find out the stable state of neural unsymmetrical growing, by using mathematical model respectively.
language of the presentation: Japanese
 
½§ùõ ÃÒ¿Î 1551054: M, 2²óÌÜȯɽ ¿ôÍý¾ðÊó³Ø ÃÓÅÄ Ï»Ê,¾¾ËÜ ·ò°ì,µ×ÊÝ ¹§ÉÙ
title: Steering behavior model based on visual information
abstract: Driving behavior prediction is important to an advanced driver assistance system (ADAS) and automatic driving system. A driver control the steering wheel using visual information. We focused on optical flow and that predicted steering behaviors. We conducted experiments to get data and evaluate the prediction model. As a result, we confirmed optical flow is usefull to predict steering behaviors.
language of the presentation: Japanese