-
Ryo HIRAI, Tadahiro FUJIMOTO
Session ID: 22-01-01
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
A method has been proposed to generate a free viewpoint video from a multi-viewpoint video captured by a camera array with cameras arranged in a 2D grid pattern and apply real-time editing to a target object in the video. The target object and its background are purely extracted without blurred color impurities of occluding objects in real time from the free viewpoint video by giving focal planes to their respective positions in the 3D space. Then, the extracted target object is interactively edited by translating, rotating, and scaling in a 2D image plane, and merged with the extracted background to obtain a new free viewpoint video. In this study, we propose a method to extend the editing from a 2D plane to a 3D space.
View full abstract
-
Shota KOGAWA, Makoto FUJISAWA, Masahiko MIKAWA
Session ID: 22-01-02
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
Fluids can be divided into two types: Newtonian fluids, which have constant viscosity, and non-Newtonian fluids, whose viscosity increases or decreases nonlinearly depending on the applied force. In this study, we focus on shear thickening fluid(dilatant fluids) whose viscosity increases as the applied force increases. We propose a new method to simulate a characteristics of dilatant fluids by reproduce the friction of fine particles in the fluid, which is the principle that generates non-Newtonianity, with friction constraint in the position-based method, instead of the method by controlling the viscosity coefficient, which has been used in previous studies.
View full abstract
-
-Estimation of sensory prediction error during steering angle operation-
Hiroyoshi KOMOBUCHI, Susumu MARUNO
Session ID: 22-01-03
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
We propose a predictive error estimating method that makes it possible to separate and associate measured values
of the steering angle to each adjusting hypothesis elements in the brain. Its effectiveness was verified through a virtual cycling
experiment in a 360-degree video projected to the ico sphere.
View full abstract
-
Satoshi Iizuka
Session ID: 22-01-04
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In this presentation, I will introduce how image editing techniques have been developed through deep learning. The methodology for image processing and editing has been actively researched in the fields of computer graphics and computer vision. In particular, deep learning with multi-layer neural networks has recently been applied to this task, enabling complicated image editing based on high-level image features, which is difficult with the previous approaches. I will first give an overview of research on image editing using deep learning and then focus on several specific research examples with more details. The presentation will also give an exposition on previous approaches without deep learning to explain how deep learning has improved these methods. In addition, current problems and future prospects in this field will be discussed.
View full abstract
-
Shunto NARITA, Tomonori IZUMI
Session ID: 22-01-05
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In order to improve material recycling of disposed electronic appliances , we aim to develop a system to analyze
and categorize electronic boards and parts utilizing cameras over conveyor belt . Besides common object recognition approach ,
here we focus on texts on boards and parts , which may give cues for faster and more accurate analysi s , and present a prototype
system and evaluation . Our system consists of text detection , region extraction and rotation , and text recognition . An
experiment shows 42% of texts are detected completely and recognized with over 80% similarity.
View full abstract
-
Yihong TANG, Tomonori IZUMI
Session ID: 22-01-06
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In order to improve material-recycling of disposed electronic appliances, we aim to develop a system to
analyze and categorize electronic boards and parts utilizing cameras over conveyor belt. We adopt deep learning
technology to recognize part images. Since popular CNN models for general object recognition are over performing,
we reduce and optimize the structure of the network to fit electronic parts. Our model classify a single part image
within about 120[μsec] with an accuracy of about 95[%].
View full abstract
-
Feifan Zhang, Yota Yamamoto, Yukinobu Taniguchi
Session ID: 22-01-07
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
The purpose of this research is to improve the accuracy of dairy cows’ individual identification by image
recognition using cow pattern features. To solve the problem of misidentification with similar patterns, we propose a method for
dairy cows’ individual identification using local feature alignment with an attention mechanism. This method improves the
accuracy of individual dairy cow identification by using local feature alignment to obtain the aligned distance between pairs of
images of dairy cows, considering and training the CNN using this distance. Furthermore, adding the attention mechanism will
also improve the accuracy of individual identification.
View full abstract
-
Shunsuke TSUKATANI, Yuuki SUGIMOTO, Naoki ITOU, Jun SHIMAMURA
Session ID: 22-01-08
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
We propose a method to detect trees for tree management by estimating trunks from tree regions composed of 3D point clouds that contain missing or noisy trees. Conventional methods have problems with detection accuracy, such as a single tree being split into multiple trees due to defects or multiple trees being detected as a single tree due to noise. The proposed method focuses on the trunk, which is the center of a tree, and improves tree detection accuracy by estimating clusters of trunks from horizontally segmented tree point clouds, removing noise, and then vertically integrating clusters based on the consistency of trunk union in consideration of defects.
View full abstract
-
Hodaka NAKAIE, Hiromi YOSHIDA
Session ID: 22-01-09
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In recent years, with the spread of VR goggles and smartphones, stereo image processing and stereoscopic vision have become more familiar. Posterization processing is a basic image processing that manipulates the gray level of an image and is used in many applications such as style conversion and image information adjustment. In this research, when changing the gray level of a stereo image by posterization processing, we investigated how it affects the parallax accuracy of the stereo image using the parallax information evaluation image.
View full abstract
-
Masato SHIBASAKI, Hiromi YOSHIDA
Session ID: 22-01-10
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In recent years, with the spread of VR goggles and smartphones, the opportunities to view and process stereo
images have been increasing. Accordingly, anonymization of personal information, such as mosaicing and blurring, is also
required for stereo images. However, when stereo images are anonymized by conventional techniques, the stereoscopic effect
is not maintained due to the degradation of disparity information. In this study, we propose a method to reduce the negative
effects on stereoscopic perception by performing anonymization while evaluating disparity information of stereo images. We
also confirm the effectiveness of the proposed method through quantitative evaluation using disparity information evaluation
images.
View full abstract
-
[in Japanese]
Session ID: 22-01-11
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
-
Zijun Li, Youngha Chang, Nobuhiko Mukai
Session ID: 22-01-12
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
Along with the increase in the feeding rate of the pets such as dogs and cats, the amount of time people spend together with their pets is increasing. Pets cannot show their emotions by language unlike, so we need to estimate pet’s emotions to spend comfortable times with them. In this paper, we propose a method to estimate cat’s emotion based on Action Unit defined for each part of cat’s face, which is extracted from cat images using deep learning.
View full abstract
-
Kazuki MATSUURA, Makoto FUJISAWA, Masahiko MIKAWA
Session ID: 22-01-13
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
In this paper, we proposes a system to generate environmental sounds from tree swaying based on a tree simulation. A position-based dynamics is used to simulate the real-time tree swaying, in order to obtain the timing of leaf collision and friction. For the collision sound, modal analysis is performed based on the voxel model made from a leaf shape to obtain the natural frequencies, and for the friction sound, a recursive frequency characteristic is reproduced using 1/f noise. The system can generate sound based on the simulation with various tree parameters, and reduce the time and effort required to record or synthesize sound which match computer graphics animations. As a result, the system enables simultaneous generation of both animation and sound.
View full abstract
-
−Analysis of Historical−Philosophical Development of Protestantism−
Kunio Ohno
Session ID: 22-01-14
Published: 2022
Released on J-STAGE: July 31, 2023
CONFERENCE PROCEEDINGS
RESTRICTED ACCESS
Based on intellectual curiosity and giving logicality, object analysis & design methodology of class definition
and its inheritance has been applied to philosophical method, and expanded to items information into visual
tabular format to clarify mutual relationships.
View full abstract