-
Hiroshi NAKAMURA, Ken TSUTSUGUCHI
Session ID: 19-03-01
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
We report a method to generate texture images from the coordinates and color data of a plane structure of 3D point
cloud data, by adding color information to the pixels corresponding to the three-dimensional position of the point. We also report
the result of evaluating the image legibility comparing generated image and processed images.
View full abstract
-
Kenta IRIE, Shunichi YONEMURA, Ken TSUTSUGUCHI
Session ID: 19-03-02
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In order to summarize the sign language video temporally, we aim to create a technology that automatically
extracts keyframe images from the video. Based on the hypothesis that the keyframes exist where the hand or finger
trajectories take extreme values in time, we extracted the movement of the hand from the sign language video using OpenPose
and approximated the trajectory with smooth curves. By doing so, we reduce the number of keyframe candidates. This method
is expected to generate efficient keyframe images.
View full abstract
-
Motoki KOKABU, Sho OOI, Mutsuo SANO
Session ID: 19-03-03
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In recent years, the animation industry uses many 3DCG models to improve quality and production efficiency.
However, creating the 3D model needs special skills and knowledge. A method of creating the 3D model is a structure from
motion (SfM) technology, but illustrator's work increase because that technology is necessary multiple perspectives images and
many images. In this paper, we learn a feature from many illustration images, consider to create the 3D model from illustration
image using obtained features by deep learning.
View full abstract
-
Takumi NAKAMURA, Ken TSUTSUGUCHI, Isamu IGARASHI, Shingo ANDO
Session ID: 19-03-04
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
We propose a method of controlling home appliances by head movement of a person wearing an eyeglass type
camera. The IoT light is controlled by the direction of viewing the digital watermarked image and the movement of the camera.
We made a prototype system and report on usability evaluation using subjects.
View full abstract
-
Ayumi FUJIWARA, Shinji IKARUGA, Takeshi KOBAYASHI, Katsumi TADAMURA
Session ID: 19-03-05
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
The number of foreign tourists increases year by year and impact of economy brought with consumption of
them is so large that invitation of the foreign tourist has been enthusiastically promoted all over Japan. On the other hand,
preparation against the rapid increase of them is not enough. Therefore providing environment which ensure that every foreign
visitor can enjoy traveling comfortably is the urgent challenge. In particular, the multilingualization of the guidance notation of
the station and the bus stop in a rural city is insufficient, and this becomes one of causes of difficulty of the use of the public
transportation system by the foreign tourist. As a solution of this problem, a software which plays a role as a guide from getting
on a train/bus to getting off it at an unmanned station and a bus stop in a rural city, assuming use by a foreign tourist whose
Japanese ability is insufficient, is proposed in this paper. More specifically, a prototype of software providing necessary
information using AR technology and GPS information at an unmanned railroad station and a bus stop was developed and its
usefulness was evaluated by real use of an unmanned station and bus service
View full abstract
-
Atsushi IMURA , Kazuyuki TAKEMAE, Hiroo AZEGAMI, Munetoshi IWAKIRI , K ...
Session ID: 19-03-06
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
Buried cultural properties such as archeological sites may be found in the process of conducting
development business such as civil engineering works. In such a case, we have to excavate the archeological site
under the Cultural Property Protection Act, and archeologists trace the outline of the archeological site and
record the archeological drawings for posterity. However, the creation of archeological drawings requires a lot of
time since it is performed by hands, which often hinders the construction work.Thus, it is required to improve
the efficiency of the work related to archeological drawings. In this paper, we propose a method for extracting
outlines that contribute to the creation of archeological drawings by using edge features from high-resolution
RGB-D images generated from multiple images of captured at the archeological site.
View full abstract
-
SATOSHI ONO
Session ID: 19-03-07
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
This talk focuses on Evolutionary Computation (EC) that is a black-box optimization framework for non-differentiable, globally multimodal, and/or ambiguous objective functions, and introduces optimization techniques and
our applications that take advantage of EC.
View full abstract
-
Suren Kawahara, Daisuke Takahasi, Noriyoshi Okamoto, Shin Kimura, Reik ...
Session ID: 19-03-08
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
t In order to perform keyboard input, the input method according to the home position is recommended. This is method
requires to grasp the fingering and the position of each key. However, beginners often input while visually recognizing the
keyboard, and tend to avoid fingering based on the home position, such as inputting all keys with the same finger. This paper
proposes a touch-typing learning game that always uses the index finger as a reference for the home position, and measures and
reports the effect in order to make the fingering corresponding to the input key easier to learn.
View full abstract
-
Yuji UMEMOTO, Ken-ichi TANAKA, Kohtaro KAMIZONO
Session ID: 19-03-09
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In information education, students will need to train typing skills in order to learn software operations. In this study,
we focus on Excel as software widely used in the curriculum, and consider the relationship between mastery of functions and
required typing ability. The subjects of the survey were high school students, and as a measure of Excel proficiency, the degree
of appearance of function-related answer words was used in the association method. We confirm the relationship between the
number of answer words that indicate the function related to Excel obtained from the association method and the number of
typing keystrokes per minute.
View full abstract
-
-Experiments using R-G Neutral- and Y-B Only Changed-Stimuli or Vice Versa-
Rei Endo, Kanta Mochihara, Shoko Hira, Minoru Ohkoba, Tomoharu Ishika ...
Session ID: 19-03-10
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
Immanent dichromaticity in trichromatic observers by using MDS (Multidimensional-Scaling) are investigated.
The results show that color-constellations yielded when observing R-G neutral- and Y-B only changed-stimuli provides strong
evidence of concave-shaped dichromic in case of popular Japanese with trichromatic color vision. The results of additional
paired comparison tests on subjective saliency reveal that the second coordinate of MDS reflects perceived chromatic saliency.
In other words, it means that color normal observers also have evaluation axis based on reliability delivered from deviations in
short- or long-wave length of perceived light sources. It is quite reasonable that the human visual system adapted to natural
environment, especially rural areas. On the other hand, the results of paired comparison tests yielded when observing Y-B
neutral- and R-G only changed-stimuli provides strong evidence that the second coordinate of MDS reflects perceived
achromatic saliency
View full abstract
-
Keiju HIRATA, Jun Ohya, Ken MASAMUNE
Session ID: 19-03-11
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In this paper, we deal with the problem of recognizing the surgical procedure from the image acquired by the
camera placed in an operating room camera. To solve this problem, we propose a method for recognizing surgical steps by
detecting joints of the surgeons and nurses as feature points in each frame of the video sequence and classifying the surgical
steps using a Bag of Visual Words, which creates histograms of the feature points. The feature points of the multiple persons
are tracked using a transfer learning of Bottom-Up type learning model that are trained using a small number of data. Then, the
detected feature points are connected so that a skeleton model is depicted in the, and the detected feature points are clustered in
the feature space so that a classifier is created. Using the classifier, the surgical step is recognized. Experiments on
recognizing surgical steps using the video sequence acquired by the camera placed in the operating room in which two
surgeons conduct a mock surgery are conducted. The experiments show the validity of the proposed method.
View full abstract
-
Kazuki HASHIMOTO, Sho OOI, Haruo NOMA, Mutsuo SANO, Satoshi UMEDA, Haj ...
Session ID: 19-03-12
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
Patients with acquired brain injury (dementia and mild cognitive impairment: MCI) have disorders such as
an executive function, memory function, and attention function. The causes of disability are traffic accidents and illness.
It is important to raise awareness of the patient as rehabilitation of acquired brain injury (cognitive rehabilitation) task.
For that purpose, the system needs to present the patient with the experienced video and quantified cognitive
information. However, conventional tests use special tools and its test is different behavior of daily living. Therefore, the
subject is a load to take the test and a doctor cannot understand the cognitive state in real-time. From the above
background, we aim to evaluate cognitive function dynamically and in real-time from cooking behavior in activities of
daily living. Specifically, our system classified the danger level during cooking by using results of skeleton recognition
and object recognition for the action of handling kitchen knives in cooking behavior.
View full abstract
-
Takeshi Morishita, Hiroshi Nagahashi, Jun Ohya
Session ID: 19-03-13
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
Guide dogs are effective for guiding visually impaired people who walk outdoors, but there is a problem that a
sufficient number of guide dogs are not available. The authors are conducting research on drones that can serve as guide dogs.
Such drones need to achieve various functions. In this paper, we focus on “walking on the edge of the road”, which is one of
the roles of guide dogs. This paper proposes a method for estimating the position of the visually impaired and the walking
area from the images acquired by the camera attached to the drone. For estimating the position of the visually impaired, the
two-dimensional coordinates of the person in the camera image are obtained using the Single Shot Multibox Detector, a
general object detection method. Next, the position of a visually impaired person is estimated by transforming two-dimensional
coordinates to three-dimensional coordinates using a pinhole camera model. For the estimation of the walking area, the curb
and white line in the road are detected by combining the segmentation of the road surface area with the edge detection and
Hough transform. An experiment was performed for estimating the curb and white line and the position of the visually
impaired. The experimental results show the validity of the proposed method.
View full abstract
-
-User Interfaces Employing Magnifier-
Shogo Hino, Shoko Hira, Sakuichi Ohtsuka
Session ID: 19-03-14
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
An aiding tool of viewing PC display, which is employed to prevent overlook left side of the focusing window, for unilateral
spatial neglect observers, is investigated. Basic concept of this tool is magnifying a part of display based on observer's requests. As the
results of user-interface examinations, three points are employed: (1) an whole window flushing effect in order to paying easy attention when
opening the magnifying window, (2) temporal control sequence of typing double same special key (Ctrl, Alt, etc.) and subsequent another
normal key for easy operation by hemiplegia users, and (3) a combination color design adopting to four sides of a window frame for easy
paying attention to the left side. This tool is also effective to users with presbyopia and/or users who want to add more active window effects
for presentations.
View full abstract
-
Hiroshi UNNO, Kazutake UEHIRA
Session ID: 19-03-15
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
This paper describes a method that transmits invisibly attached multiple information to displayed video image on flat
panel display using time division multiple transmission system. We proposed information format for video image displayed on
the display. We also proposed and examined a method that enable us to extract and select any kind of information which invisibly
attached to displayed image captured with a video camera. We demonstrated the feasibility of the method we propose.
View full abstract
-
Takanori HIGAKI, Yoshio YANAGIHARA, Koji KINOSHITA, Masaharu ISSHIKI
Session ID: 19-03-16
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
There is a problem that counterfeit bills are rampant in China, and it is necessary to take countermeasures. Therefore,
in order to properly manage bills, it is possible to track bills and judge fake bills by recording and sharing the serial number
described. Therefore, a system that can identify serial numbers from bills with high accuracy is needed. In the conventional
system, character identification is performed using CNN, but high-precision character identification is performed using BNN,
which can perform higher-speed processing due to problems such as speed. However, the accuracy of BNN decreases because
the product weight is calculated by replacing the connection weight of the CNN and the value of the feature map with -1 or 1.
In this paper, we aim to improve the accuracy by using the error diffusion method for the error in binarization.
View full abstract
-
Shoto SEKI, Masashi KAMEDA
Session ID: 19-03-17
Published: 2020
Released on J-STAGE: August 31, 2021
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
We have developed a reconstructive image coding method using wavelet transform
for any high resolution images. In order to improve the image quality in the conventional method,
this paper presents two kinds of generation methods of high frequency components, and the
encoded image with the optimum selection of the generated high frequency components using the
original image is obtained. Also, some characteristics of the appropriate high frequency
components are shown in our experiments.
View full abstract