Misc - AMANO Toshiyuki

画像の学習に基づく姿勢推定手法の学習サンプル数と精度の関係性の検討

奥川裕之, 原田健吾, 玉木徹, 天野敏之, 金田和文

電気・情報関連学会中国支部連合大会講演論文集(CD−ROM) ( 電気・情報関連学会中国支部 )  59th ( 59 ) 470 - 470   2008.10

Linear 3D object pose estimation with dense sample images

玉木 徹, 天野 敏之

MIRU 2008 画像の認識・理解シンポジウムダイジェスト     501 - 501   2008.07

　View Summary

In the image parameter estimation by the linear regression, it has very high degrees of freedom for the decision of regression coefficients, because the dimension of image vector is huge high. In this paper, we propose a sequential regression coefficient calculation algorithm, and we realize its calculation for dense samples with reasonable computational cost. Moreover, we apply this method to the pose estimation of the 3-D object, and we discuss about limit of parameter estimation ability by the linear regression with the coil-20 image library.

Representing pose by spherical functions for pose estimation

玉木 徹, 天野 敏之, 金田 和文

MIRU 2008 画像の認識・理解シンポジウムダイジェスト     502 - 502   2008.07

　View Summary

We propose a novel representation of pose in 3 degrees-of-freedom (DOF) for view-based pose estimation. First we show that a conventional representation of pose in 1 DOF is a Fourier basis, and extend the observation to 2 DOF with spherical harmonics. Then we represent 3 DOF pose with spherical functions that are continuous on SO(3), and give transformations from the spherical functions representation to a quaternion and a rotation matrix.

Expanding images of an object under 3DOF rotation with spherical functions and its applications to image interpolation and pose estimation

玉木 徹, 天野 敏之, 金田 和文

MIRU 2008 画像の認識・理解シンポジウム論文集     1169 - 1174   2008.07

　View Summary

We propose an implementation for expanding images of an object under 3 DOF rotation with spherical functions, an orthonormal basis on SO(3), and describe its applications to image interpolation and pose estimation.

Representing pose by spherical functions for pose estimation

玉木 徹, 天野 敏之, 金田 和文

MIRU 2008 画像の認識・理解シンポジウム論文集     1134 - 1141   2008.07

　View Summary

We propose a novel representation of pose in 3 degrees-of-freedom (DOF) for view-based pose estimation. First we show that a conventional representation of pose in 1 DOF is a Fourier basis, and extend the observation to 2 DOF with spherical harmonics. Then we represent 3 DOF pose with spherical functions that are continuous on SO(3), and give transformations from the spherical functions representation to a quaternion and a rotation matrix.

Expanding images of an object under 3DOF rotation with spherical functions and its applications to image interpolation and pose estimation

玉木 徹, 天野 敏之, 金田 和文

MIRU 2008 画像の認識・理解シンポジウムダイジェスト     507 - 507   2008.07

　View Summary

We propose an implementation for expanding images of an object under 3 DOF rotation with spherical functions, an orthonormal basis on SO(3), and describe its applications to image interpolation and pose estimation.

A Tracking Method using Multi-core CPUs for Virtual 3D Pop-up Books

TAMURA Ryosuke, AMANO Toshiyuki, KATO Hirokazu

IPSJ SIG Notes. CVIM ( Information Processing Society of Japan (IPSJ) )  2008 ( 3 ) 341 - 342   2008.01

　View Summary

Virtual pop-up books are systems which can overlay a captured image with 3D CG objects based on augmented reality technology. 3D geometrical consistency in the composite image can be kept by estimating the relative pose of the camera from the physical book. We have been studying about this pose estimation method using feature tracking based on template matching. Since multi-core CPUs have been spread in recent years, it can be easily achieved to improve the performance speed by parallel processing on a standard PC environment. We modified our tracking codes for parallel processing by multi-core CPUs and improved the performance speed and robustness of the tracking.

A Tracking Method using Multi-core CPUs for Virtual 3D Pop-up Books

TAMURA Ryosuke, AMANO Toshiyuki, KATO Hirokazu

Technical report of IEICE. PRMU ( The Institute of Electronics, Information and Communication Engineers )  107 ( 427 ) 121 - 122   2008.01

　View Summary

Virtual pop-up books are systems which can overlay a captured image with 3D CG objects based on augmented reality technology. 3D geometrical consistency in the composite image can be kept by estimating the relative pose of the camera from the physical book. We have been studying about this pose estimation method using feature tracking based on template matching. Since multi-core CPUs have been spread in recent years, it can be easily achieved to improve the performance speed by parallel processing on a standard PC environment. We modified our tracking codes for parallel processing by multi-core CPUs and improved the performance speed and robustness of the tracking.

PRMU Algorithm Contest 2007 "Let's Make Jigsaw Puzzle! -Image Reconstruction by Matching-" Summary Report and Prize Winning Algorithms

AMANO Toshiyuki, TAMAKI Toru, MURAKAMI Takenori, AOKI Yasuaki, BITO Koji, AKIMA Yuuta, ARAKI Kousuke, SAKO MASASHI, HANAOKA Shouhei

IEICE technical report ( The Institute of Electronics, Information and Communication Engineers )  107 ( 384 ) 125 - 136   2007.12

　View Summary

This article gives a summary report on the algorithm contest titled "Let's Make Jigsaw Puzzle! -Image Reconstruction by Matching-", which is sponsored by PRMU (Pattern Recognition and Media Understanding) Technical Group of IEICE-ISS, with brief overviews of prize-winning algorithms provided by the winners. The contest in this year asked for the algorithms that solve the arrangement of pieces for the image reconstruction. The total number of applications is 84, and the brilliant 7 algorithms have won the prizes.

PRMU Algorithm Contest 2007 "Let's Make Jigsaw Puzzle! -Image Reconstruction by Matching-" Summary Report and Prize Winning Algorithms

AMANO Toshiyuki, TAMAKI Toru, MURAKAMI Takenori, AOKI Yasuaki, BITO Koji, AKIMA Yuuta, ARAKI Kousuke, SAKO MASASHI, HANAOKA Shouhei

IEICE technical report ( The Institute of Electronics, Information and Communication Engineers )  107 ( 384 ) 125 - 136   2007.12

　View Summary

This article gives a summary report on the algorithm contest titled "Let's Make Jigsaw Puzzle! -Image Reconstruction by Matching-", which is sponsored by PRMU (Pattern Recognition and Media Understanding) Technical Group of IEICE-ISS, with brief overviews of prize-winning algorithms provided by the winners. The contest in this year asked for the algorithms that solve the arrangement of pieces for the image reconstruction. The total number of applications is 84, and the brilliant 7 algorithms have won the prizes.

On methods for taking pedstrians' paths for prediction

篠村 祐司, 原野 友敬, 玉木 徹, 天野 敏之, 金田 和文, 内田 誠一

第9回IEEE広島支部学生シンポジウム(HISS)論文集 ( IEEE広島支部学生シンポジウム(HISS) )    2007.11

Defect Detection Using k-means Clustering and Eigenspace Outlier Detector

AMANO Toshiyuki

IEICE technical report ( The Institute of Electronics, Information and Communication Engineers )  107 ( 281 ) 111 - 116   2007.10

　View Summary

In this paper, a new approach of defect detection from complex texture patterns that have structural regularity for automated visual inspection is proposed. The approach uses correlation based defect detection, and any kind of model of defect feature is unnecessary. Additionally, proposed method applies k-means clustering to the samples acquired by pixel-by-pixel movement of sampling window, and achieves a very sensitive defect detection. In the experiment, the effectiveness is shown through the experiment by the complex texture patterns that have structural regularity.

コンパクトな辞書データによる高速な3次元物体認識と姿勢推定

木村将之, 天野敏之, 玉木徹, 梅崎太造

電気関係学会東海支部連合大会講演論文集(CD−ROM)   2007   ROMBUNNO.O-387   2007.09

On the complex and quaternion subspaces for view-based pose estimation

TAMAKI TORU, AMANO TOSHIYUKI, KANEDA KAZUFUMI, ICHIHARA YUMIKO

IPSJ SIG Notes. CVIM ( Information Processing Society of Japan (IPSJ) )  160 ( 87 ) 181 - 186   2007.09

　View Summary

This paper proposes a new method for construction of a complex subspace for rotation about a single axis and a quaternion subspace for 3DOF rotation of a 3D object, then demonstrates with the proposed method view-based pose estimation.

On the complex and quaternion subspaces for view-based pose estimation

TAMAKI TORU, AMANO TOSHIYUKI, KANEDA KAZUFUMI, ICHIHARA YUMIKO

IEICE technical report ( The Institute of Electronics, Information and Communication Engineers )  107 ( 207 ) 181 - 186   2007.08

　View Summary

This paper proposes a new method for construction of a complex subspace for rotation about a single axis and a quaternion subspace for 3DOF rotation of a 3D object, then demonstrates with the proposed method view-based pose estimation.

Projecting images of rotating object onto complex subspace by the matrix of cyclic group and its application

玉木 徹, 天野 敏之, 金田 和文

MIRU2007 画像の認識・理解シンポジウム ダイジェスト ( 画像情報フォーラム )    563 - 563   2007.07

Projecting images of rotating object onto subspace by block diagonalization of the matrix of cyclic group and its application

玉木 徹, 天野 敏之, 金田 和文

MIRU2007 画像の認識・理解シンポジウム ダイジェスト ( 画像情報フォーラム )    562 - 562   2007.07

Projecting images of rotating object onto subspace by block diagonalization of the matrix of cyclic group and its application

玉木 徹, 天野 敏之, 金田 和文

MIRU 2007 画像の認識・理解シンポジウム論文集 ( 画像情報フォーラム )    982 - 987   2007.07

　View Summary

This paper proposes a linear map that projects images of an rotating object about an axis onto a subspace where the rotation of the object is represented with many simple rotations in two dimensional subspaces. First we consider a cyclic group that permutates the image sequence, then construct a matrix of the cyclic group by multiplication of low rank matrices with block diagonalization. We demonstrate interpolation of the image sequence by rotation in the subspace, and discuss the ability of the proposed linear map onto the subspace.

Prediction of a path of pedestrian using combination of Eigenspace and null space

篠村 祐司, 玉木 徹, 天野 敏之, 金田 和文

MIRU 2007 画像の認識・理解シンポジウム論文集 ( 画像情報フォーラム )    774 - 779   2007.07

　View Summary

This paper proposes a method for Eigenspace-based prediction of a person's path in future with current path. This method uses projection and back projection of the current path onto Eigenspace. However, the prediction does not take the nature of a path of pedestrian and results in non-smooth path. The proposed method modify the prediction by using null vectors in the orthocomplement of the learned Eigenspace and adding linear combinations of the null vectors to the prediction so that the path becomes smooth. Coe?cients of the linear combinations are computed by decent gradient method. Some experimental results on actual pathes are shown.

Projecting images of rotating object onto complex subspace by the matrix of cyclic group and its application

玉木 徹, 天野 敏之, 金田 和文

MIRU 2007 画像の認識・理解シンポジウム論文集 ( 画像情報フォーラム )    988 - 993   2007.07

　View Summary

This paper proposes a linear map that projects images of an rotating object about anax is onto a complex subspace where the rotation is represented with an simple rotation in one dimensional complex space. First we consider a cyclic group that permutates the image sequence, then construct a complex matrix of the cyclic group by multiplication of low rank matrices. We demonstrate interpolation of the image sequence by rotation in the subspace ,and discuss the ability of the proposed linear map onto the complex subspace.